Bump version

* Add a test for generators

* Remove output annotations from decorators

Also guarded torch imports for better compatibility with
requirements.txt

* Add flag to the main meta class to skip the typecheck

* Return to the old solution

* Make async tests work

* Minor adjustments/fixing typos

* Correct Python path for new tests

* Remove some jax-dependent code

* Implement equality for MetaArrays

* Make all Dim variations frozen dataclasses

* Shorten AbstractArray methods

* Final touches

* Removing get_origin use

* Update tests with @jaxtyp

.pre-commit-config.yaml

@@ -18,11 +18,11 @@
 # CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 repos:
-  - repo: https://github.com/ambv/black
-    rev: 22.3.0
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.1.7
     hooks:
-    - id: black
-  - repo: https://github.com/charliermarsh/ruff-pre-commit
-    rev: 'v0.0.255'
-    hooks:
-      - id: ruff
+      - id: ruff  # linter
+        types_or: [ python, pyi, jupyter ]
+        args: [ --fix ]
+      - id: ruff-format  # formatter
+        types_or: [ python, pyi, jupyter ]

README.md

@@ -10,7 +10,7 @@ Type annotations **and runtime type-checking** for:
 ```python
 from jaxtyping import Array, Float, PyTree
 
-# Accepts floating-point 2D arrays with matching dimensions
+# Accepts floating-point 2D arrays with matching axes
 def matrix_multiply(x: Float[Array, "dim1 dim2"],
                     y: Float[Array, "dim2 dim3"]
                   ) -> Float[Array, "dim1 dim3"]:
@@ -49,14 +49,22 @@ Available at [https://docs.kidger.site/jaxtyping](https://docs.kidger.site/jaxty
 
 [Diffrax](https://github.com/patrick-kidger/diffrax): numerical differential equation solvers.
 
-[Lineax](https://github.com/google/lineax): linear solvers and linear least squares.
+[Optimistix](https://github.com/patrick-kidger/optimistix): root finding, minimisation, fixed points, and least squares.
 
-[Eqxvision](https://github.com/paganpasta/eqxvision): computer vision models.
+[Lineax](https://github.com/google/lineax): linear solvers.
+
+[BlackJAX](https://github.com/blackjax-devs/blackjax): probabilistic+Bayesian sampling.
+
+[Orbax](https://github.com/google/orbax): checkpointing (async/multi-host/multi-device).
 
 [sympy2jax](https://github.com/google/sympy2jax): SymPy<->JAX conversion; train symbolic expressions via gradient descent.
 
+[Eqxvision](https://github.com/paganpasta/eqxvision): computer vision models.
+
 [Levanter](https://github.com/stanford-crfm/levanter): scalable+reliable training of foundation models (e.g. LLMs).
 
+[PySR](https://github.com/milesCranmer/PySR): symbolic regression. (Non-JAX honourable mention!)
+
 ### Disclaimer
 
 This is not an official Google product.

docs/api/advanced-features.md

@@ -7,6 +7,10 @@
         members:
             false
 
+## Printing axis bindings
+
+::: jaxtyping.print_bindings
+
 ## Introspection
 
 If you're writing your own type hint parser, then you may wish to detect if some Python object is a jaxtyping-provided type.

docs/api/array.md

@@ -10,8 +10,9 @@ The shape should be a string of space-separated symbols, such as `"a b c d"`. Ea
 
 - `int`: fixed-size axis, e.g. `"28 28"`.
 - `str`: variable-size axis, e.g. `"channels"`.
-- A symbolic expression (without spaces!) in terms of other variable-size axes, e.g.  
-    `def remove_last(x: Float[Array, "dim"]) -> Float[Array, "dim-1"]`.
+- A symbolic expression in terms of other variable-size axes, e.g.  
+    `def remove_last(x: Float[Array, "dim"]) -> Float[Array, "dim-1"]`.  
+    Symbolic expressions must not use any spaces, otherwise each piece is treated as as a separate axis.
 
 When calling a function, variable-size axes and symbolic axes will be matched up across all arguments and checked for consistency. (See [Runtime type checking](./runtime-type-checking.md).)
 
@@ -19,11 +20,12 @@ When calling a function, variable-size axes and symbolic axes will be matched up
 
 In addition some modifiers can be applied:
 
-- Prepend `*` to a dimension to indicate that it can match multiple axes, e.g. `"*batch c h w"` will match zero or more batch axes.
-- Prepend `#` to a dimension to indicate that it can be that size *or* equal to one -- i.e. broadcasting is acceptable, e.g.  
+- Prepend `*` to an axis to indicate that it can match multiple axes, e.g. `"*batch"` will match zero or more batch axes.
+- Prepend `#` to an axis to indicate that it can be that size *or* equal to one -- i.e. broadcasting is acceptable, e.g.  
     `def add(x: Float[Array, "#foo"], y: Float[Array, "#foo"]) -> Float[Array, "#foo"]`.
-- Prepend `_` to a dimension to disable any runtime checking of that dimension (so that it can be used just as documentation). This can also be used as just `_` on its own: e.g. `"b c _ _"`.
+- Prepend `_` to an axis to disable any runtime checking of that axis (so that it can be used just as documentation). This can also be used as just `_` on its own: e.g. `"b c _ _"`.
 - Documentation-only names (i.e. they're ignored by jaxtyping) can be handled by prepending a name followed by `=` e.g. `Float[Array, "rows=4 cols=3"]`.
+- Prepend `?` to an axis to indicate that its size can vary within a PyTree structure. (See [PyTree annotations](./pytree.md).)
 
 When using multiple modifiers, their order does not matter.
 
@@ -36,28 +38,38 @@ As a special case:
 - To denote a scalar shape use `""`, e.g. `Float[Array, ""]`.
 - To denote an arbitrary shape (and only check dtype) use `"..."`, e.g. `Float[Array, "..."]`.
 - You cannot have more than one use of multiple-axes, i.e. you can only use `...` or `*name` at most once in each array.
-- An example of broadcasting multiple dimensions:  
-    `def add(x: Float[Array, "*#foo"], y: Float[Array, "*#foo"]) -> Float[Array, "*#foo"]`.
 - A symbolic expression cannot be evaluated unless all of the axes sizes it refers to have already been processed. In practice this usually means that they should only be used in annotations for the return type, and only use axes declared in the arguments.
+- Symbolic expressions are evaluated in two stages: they are first evaluated as f-strings using the arguments of the function, and second are evaluated using the processed axis sizes. The f-string evaluation means that they can use local variables by enclosing them with curly braces, e.g. `{variable}`, e.g.
+    ```python
+    def full(size: int, fill: float) -> Float[Array, "{size}"]:
+        return jax.numpy.full((size,), fill)
+
+    class SomeClass:
+        some_value = 5
+
+        def full(self, fill: float) -> Float[Array, "{self.some_value}+3"]:
+            return jax.numpy.full((self.some_value + 3,), fill)
+    ```
 
 ## Dtype
 
 The dtype should be any one of (all imported from `jaxtyping`):
 
 - Any dtype at all: `Shaped`
-  - Boolean: `Bool`
-  - PRNG key: `Key`
-  - Any integer, unsigned integer, floating, or complex: `Num`
-    - Any floating or complex: `Inexact`
-      - Any floating point: `Float`
-        - Of particular precision: `BFloat16`, `Float16`, `Float32`, `Float64`
-      - Any complex: `Complex`
-        - Of particular precision: `Complex64`, `Complex128`
-    - Any integer or unsigned intger: `Integer`
-      - Any unsigned integer: `UInt`
-        - Of particular precision: `UInt8`, `UInt16`, `UInt32`, `UInt64`
-      - Any signed integer: `Int`
-        - Of particular precision: `Int8`, `Int16`, `Int32`, `Int64`
+    - Boolean: `Bool`
+    - PRNG key: `Key`
+    - Any integer, unsigned integer, floating, or complex: `Num`
+        - Any floating or complex: `Inexact`
+            - Any floating point: `Float`
+                - Of particular precision: `BFloat16`, `Float16`, `Float32`, `Float64`
+            - Any complex: `Complex`
+                - Of particular precision: `Complex64`, `Complex128`
+        - Any integer or unsigned intger: `Integer`
+            - Any unsigned integer: `UInt`
+                - Of particular precision: `UInt4`, `UInt8`, `UInt16`, `UInt32`, `UInt64`
+            - Any signed integer: `Int`
+                - Of particular precision: `Int4`, `Int8`, `Int16`, `Int32`, `Int64`
+        - Any floating, integer, or unsigned integer: `Real`.
 
 Unless you really want to force a particular precision, then for most applications you should probably allow any floating-point, any integer, etc. That is, use
 ```python
@@ -95,16 +107,15 @@ BatchImage = Float[Array, "batch channels height width"]
 
 Note that `jaxtyping.{Array, ArrayLike}` are only available if JAX has been installed.
 
-## Scalars, PRNGKeys
+## Scalars, PRNG keys
 
 For convenience, jaxtyping also includes `jaxtyping.Scalar`, `jaxtyping.ScalarLike`, and `jaxtyping.PRNGKeyArray`, defined as:
 ```python
 Scalar = Shaped[Array, ""]
 ScalarLike = Shaped[ArrayLike, ""]
 
-# Depending on the value of `JAX_ENABLE_CUSTOM_PRNG`:
-PRNGKeyArray = Key[Array, ""]
-PRNGKeyArray = UInt32[Array, "2"]
+# Left: new-style typed keys; right: old-style keys. See JEP 9263.
+PRNGKeyArray = Union[Key[Array, ""], UInt32[Array, "2"]]
 ```
 
 Recalling that shape-and-dtype specified jaxtyping arrays can be nested, this means that e.g. you can annotate the output of `jax.random.split` with `Shaped[PRNGKeyArray, "2"]`, or e.g. an integer scalar with `Int[Scalar, ""]`.

docs/api/pytree.md

@@ -11,4 +11,36 @@
 
 ---
 
+## Path-dependent shapes
+
+The prefix `?` may be used to indicate that the axis size can depend on which leaf of a PyTree the array is at. For example:
+```python
+def f(
+    x: PyTree[Shaped[Array, "?foo"], "T"],
+    y: PyTree[Shaped[Array, "?foo"], "T"],
+):
+    pass
+```
+The above demands that `x` and `y` have matching PyTree structures (due to the `T` annotation), and that their leaves must all be one-dimensional arrays, *and that the corresponding pairs of leaves in `x` and `y` must have the same size as each other*.
+
+Thus the following is allowed:
+```python
+x0 = jnp.arange(3)
+x1 = jnp.arange(5)
+
+y0 = jnp.arange(3) + 1
+y1 = jnp.arange(5) + 1
+
+f((x0, x1), (y0, y1))  # x0 matches y0, and x1 matches y1. All good!
+```
+
+But this is not:
+```python
+f((x1, x1), (y0, y1))  # x1 does not have a size matching y0!
+```
+
+Internally, all that is happening is that `foo` is replaced with `0foo` for the first leaf, `1foo` for the next leaf, etc., so that each leaf gets a unique version of the name.
+
+---
+
 Note that `jaxtyping.{PyTree, PyTreeDef}` are only available if JAX has been installed.

docs/api/runtime-type-checking.md

@@ -2,21 +2,54 @@
 
 (See the [FAQ](../faq.md) for details on static type checking.)
 
-Runtime type checking **synergises beautifully with `jax.jit`!** All shape checks will be performed at trace-time only, and will not impact runtime performance.
+Runtime type checking **synergises beautifully with `jax.jit`!** All shape checks will be performed only whilst tracing, and will not impact runtime performance.
 
-Runtime type-checking should be performed using a library like [typeguard](https://github.com/agronholm/typeguard) or [beartype](https://github.com/beartype/beartype).
+There are two approaches: either use [`jaxtyping.jaxtyped`][] to typecheck a single function, or [`jaxtyping.install_import_hook`][] to typecheck a whole codebase.
 
-The types provided by `jaxtyping`, e.g. `Float[Array, "batch channels"]`, are all compatible with `isinstance` checks, e.g. `isinstance(x, Float[Array, "batch channels"])`. This means that jaxtyping should be compatible with all runtime type checkers out-of-the-box.
+In either case, the actual business of checking types is performed with the help of a runtime type-checking library. The two most popular are [beartype](https://github.com/beartype/beartype) and [typeguard](https://github.com/agronholm/typeguard). (If using typeguard, then specifically the version `2.*` series should be used. Later versions -- `3` and `4` -- have some known issues.)
 
-Some additional context is needed to ensure consistency between multiple argments (i.e. that shapes match up between arrays). For this, you can use either `jaxtyping.jaxtyped` to add this capability to a single function, or `jaxtyping.install_import_hook` to add this capability to a whole codebase. If either are too much magic for you, you can safely use neither and have just single-argument type checking.
+!!! warning
+
+    Avoid using `from __future__ import annotations`, or stringified type annotations, where possible. These are largely incompatible with runtime type checking. See also [this FAQ entry](../faq.md#dataclass-annotations-arent-being-checked-properly).
+
+---
 
 ::: jaxtyping.jaxtyped
 
 ---
 
-It can be a lot of effort to add `@jaxtyped` decorators all over your codebase.
-(Not to mention that double-decorators everywhere are a bit ugly.)
-
-The easier option is usually to use the import hook.
-
 ::: jaxtyping.install_import_hook
+
+---
+
+#### Pytest hook
+
+The import hook can be installed at test-time only, as a pytest hook. From the command line the syntax is:
+```
+pytest --jaxtyping-packages=foo,bar.baz,beartype.beartype
+```
+or in `pyproject.toml`:
+```toml
+[tool.pytest.ini_options]
+addopts = "--jaxtyping-packages=foo,bar.baz,beartype.beartype"
+```
+or in `pytest.ini`:
+```ini
+[pytest]
+addopts = --jaxtyping-packages=foo,bar.baz,beartype.beartype
+```
+This example will apply the import hook to all modules whose names start with either `foo` or `bar.baz`. The typechecker used in this example is `beartype.beartype`.
+
+#### IPython extension
+
+If you are running in an IPython environment (for example a Jupyter or Colab notebook), then the jaxtyping hook can be automatically ran via a custom magic:
+```python
+import jaxtyping
+%load_ext jaxtyping
+%jaxtyping.typechecker beartype.beartype  # or any other runtime type checker
+```
+Place this at the start of your notebook -- everything that is directly defined in the notebook, after this magic is run, will be hook'd.
+
+#### Other runtime type-checking libraries
+
+Beartype and typeguard happen to be the two most popular runtime type-checking libraries (at least at time of writing), but jaxtyping should be compatible with all runtime type checkers out-of-the-box. The runtime type-checking library just needs to provide a type-checking decorator (analgous to `beartype.beartype` or `typeguard.typechecked`), and perform `isinstance` checks against jaxtyping's types.

docs/faq.md

@@ -20,6 +20,24 @@ Some tooling in the Python ecosystem assumes that only the latter is true, and w
 
 In the case of `flake8`, or Ruff, this can be resolved. Multi-dimensional arrays (e.g. `Float32[Array, "b c"]`) will throw a very unusual error (F722, syntax error in forward annotation), so you can safely just disable this particular error globally. Uni-dimensional arrays (e.g. `Float32[Array, "x"]`) will throw an error that's actually useful (F821, undefined name), so instead of disabling this globally, you should instead prepend a space to the start of your shape, e.g. `Float32[Array, " x"]`. `jaxtyping` will treat this in the same way, whilst `flake8` will now throw an F722 error that you can disable as before.
 
+## Dataclass annotations aren't being checked properly.
+
+Stringified dataclass annotations, e.g.
+```python
+@dataclass()
+class Foo:
+   x: "int"
+```
+will be silently skipped without checking them. This is because these are essentially impossible to resolve at runtime. Such stringified annotations typically occur either when using them for forward references, or when using `from __future__ import annotations`. (You should essentially never use the latter, it is largely incompatible with runtime type checking and as such is [being replaced in Python 3.13](https://peps.python.org/pep-0649/).)
+
+Partially stringified dataclass annotations, e.g.
+```python
+@dataclass()
+class Foo:
+   x: tuple["int"]
+```
+will likely raise an error, and must not be used at all.
+
 ## Does jaxtyping use [PEP 646](https://www.python.org/dev/peps/pep-0646/) (variadic generics)?
 
 The intention of PEP 646 was to make it possible for static type checkers to perform shape checks of arrays. Unfortunately, this still isn't yet practical, so jaxtyping deliberately does not use this. (Yet?)

docs/index.md

@@ -24,7 +24,7 @@ The annotations provided by jaxtyping are compatible with runtime type-checking
 ```python
 from jaxtyping import Array, Float, PyTree
 
-# Accepts floating-point 2D arrays with matching dimensions
+# Accepts floating-point 2D arrays with matching axes
 def matrix_multiply(x: Float[Array, "dim1 dim2"],
                     y: Float[Array, "dim2 dim3"]
                   ) -> Float[Array, "dim1 dim3"]:

docs/requirements.txt

@@ -3,7 +3,7 @@ mkdocs-material==7.3.6   # Theme
 pymdown-extensions==9.4  # Markdown extensions e.g. to handle LaTeX.
 mkdocstrings==0.17.0     # Autogenerate documentation from docstrings.
 mknotebooks==0.7.1       # Turn Jupyter Lab notebooks into webpages.
-pytkdocs_tweaks==0.0.5   # Tweaks mkdocstrings to improve various aspects
+pytkdocs_tweaks==0.0.8   # Tweaks mkdocstrings to improve various aspects
 mkdocs_include_exclude_files==0.0.1  # Tweak which files are included/excluded
 jinja2==3.0.3            # Older version. After 3.1.0 seems to be incompatible with current versions of mkdocstrings.
 pygments==2.14.0

jaxtyping/__init__.py

@@ -17,58 +17,37 @@
 # IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 # CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
+import functools as ft
 import importlib.metadata
+import importlib.util
 import typing
+import warnings
+from typing import Union
 
-
-try:
-    import jax
-except ImportError:
-    has_jax = False
-else:
-    has_jax = True
-    del jax
-
-# First import some things as normal
 from ._array_types import (
     AbstractArray as AbstractArray,
     AbstractDtype as AbstractDtype,
     get_array_name_format as get_array_name_format,
     set_array_name_format as set_array_name_format,
 )
+from ._config import config as config
 from ._decorator import jaxtyped as jaxtyped
+from ._errors import (
+    AnnotationError as AnnotationError,
+    TypeCheckError as TypeCheckError,
+)
 from ._import_hook import install_import_hook as install_import_hook
+from ._ipython_extension import load_ipython_extension as load_ipython_extension
+from ._storage import print_bindings as print_bindings
 
 
-# Now import Array and ArrayLike
 if typing.TYPE_CHECKING:
-    # For imports, we need to explicitly `import X as X` in order for Pyright to see
-    # them as public. See discussion at https://github.com/microsoft/pyright/issues/2277
+    import typing_extensions
+
     from jax import Array as Array
-    from jax.typing import ArrayLike as ArrayLike
-elif has_jax:
-    if getattr(typing, "GENERATING_DOCUMENTATION", False):
+    from jax.tree_util import PyTreeDef as PyTreeDef
+    from jax.typing import ArrayLike as ArrayLike, DTypeLike as DTypeLike
 
-        class Array:
-            pass
-
-        Array.__module__ = "builtins"
-
-        class ArrayLike:
-            pass
-
-        ArrayLike.__module__ = "builtins"
-    else:
-        from jax import Array as Array
-
-        try:
-            from jax.typing import ArrayLike as ArrayLike
-        except (ModuleNotFoundError, ImportError):
-            pass
-
-
-# Import our dtypes
-if typing.TYPE_CHECKING:
     # Introduce an indirection so that we can `import X as X` to make it clear that
     # these are public.
     from ._indirection import (
@@ -83,6 +62,7 @@ if typing.TYPE_CHECKING:
         Float64 as Float64,
         Inexact as Inexact,
         Int as Int,
+        Int4 as Int4,
         Int8 as Int8,
         Int16 as Int16,
         Int32 as Int32,
@@ -90,49 +70,18 @@ if typing.TYPE_CHECKING:
         Integer as Integer,
         Key as Key,
         Num as Num,
+        PRNGKeyArray as PRNGKeyArray,
+        Real as Real,
+        Scalar as Scalar,
+        ScalarLike as ScalarLike,
         Shaped as Shaped,
         UInt as UInt,
+        UInt4 as UInt4,
         UInt8 as UInt8,
         UInt16 as UInt16,
         UInt32 as UInt32,
         UInt64 as UInt64,
     )
-else:
-    from ._array_types import (
-        BFloat16 as BFloat16,
-        Bool as Bool,
-        Complex as Complex,
-        Complex64 as Complex64,
-        Complex128 as Complex128,
-        Float as Float,
-        Float16 as Float16,
-        Float32 as Float32,
-        Float64 as Float64,
-        Inexact as Inexact,
-        Int as Int,
-        Int8 as Int8,
-        Int16 as Int16,
-        Int32 as Int32,
-        Int64 as Int64,
-        Integer as Integer,
-        Num as Num,
-        Shaped as Shaped,
-        UInt as UInt,
-        UInt8 as UInt8,
-        UInt16 as UInt16,
-        UInt32 as UInt32,
-        UInt64 as UInt64,
-    )
-
-    if has_jax:
-        from ._array_types import Key as Key
-
-
-# Now import PyTreeDef and PyTree
-if typing.TYPE_CHECKING:
-    import typing_extensions
-
-    from jax.tree_util import PyTreeDef as PyTreeDef
 
     # Set up to deliberately confuse a static type checker.
     PyTree: typing_extensions.TypeAlias = getattr(typing, "foo" + "bar")
@@ -152,55 +101,145 @@ if typing.TYPE_CHECKING:
     # If they can't figure out what a type is, then they just give up and allow
     # anything. (I believe this is sometimes called `Unknown`.) Thus, this odd-looking
     # annotation, which static type checkers aren't smart enough to resolve.
-elif has_jax:
-    if hasattr(typing, "GENERATING_DOCUMENTATION"):
-        # Most parts of the Equinox ecosystem have
-        # `typing.GENERATING_DOCUMENTATION = True` when generating documentation, to
-        # add whatever shims are necessary to get pretty docs. E.g. to have type
-        # annotations appear as just `PyTree`, not `jaxtyping.PyTree`.
-        #
-        # As jaxtyping actually wants things to appear as e.g. `jaxtyping.PyTree`,
-        # rather than just `PyTree`, then it sets
-        # `typing.GENERATING_DOCUMENTATION = False`, to disable these shims.
-        #
-        # Here we do only a `hasattr` check, as we want to get this version of
-        # `PyTreeDef` in both the jaxtyping and the Equinox(/etc.) docs.
+else:
+    from ._array_types import (
+        BFloat16 as BFloat16,
+        Bool as Bool,
+        Complex as Complex,
+        Complex64 as Complex64,
+        Complex128 as Complex128,
+        Float as Float,
+        Float16 as Float16,
+        Float32 as Float32,
+        Float64 as Float64,
+        Inexact as Inexact,
+        Int as Int,
+        Int4 as Int4,
+        Int8 as Int8,
+        Int16 as Int16,
+        Int32 as Int32,
+        Int64 as Int64,
+        Integer as Integer,
+        Key as Key,
+        Num as Num,
+        Real as Real,
+        Shaped as Shaped,
+        UInt as UInt,
+        UInt4 as UInt4,
+        UInt8 as UInt8,
+        UInt16 as UInt16,
+        UInt32 as UInt32,
+        UInt64 as UInt64,
+    )
 
-        class PyTreeDef:
-            """Alias for `jax.tree_util.PyTreeDef`, which is the type of the return
-            from `jax.tree_util.tree_structure(...)`.
-            """
+    # But crucially, does not actually import jax at all. We do that dynamically in
+    # __getattr__ if required. See #178.
+    if importlib.util.find_spec("jax") is not None:
 
-        if typing.GENERATING_DOCUMENTATION:
-            # Equinox etc. docs get just `PyTreeDef`.
-            # jaxtyping docs get `jaxtyping.PyTreeDef`.
-            PyTreeDef.__module__ = "builtins"
+        @ft.cache
+        def __getattr__(item):
+            if item == "Array":
+                if getattr(typing, "GENERATING_DOCUMENTATION", False):
 
+                    class Array:
+                        pass
+
+                    Array.__module__ = "builtins"
+                    return Array
+                else:
+                    import jax
+
+                    return jax.Array
+            elif item == "ArrayLike":
+                if getattr(typing, "GENERATING_DOCUMENTATION", False):
+
+                    class ArrayLike:
+                        pass
+
+                    ArrayLike.__module__ = "builtins"
+                    return ArrayLike
+                else:
+                    import jax.typing
+
+                    return jax.typing.ArrayLike
+            elif item == "PRNGKeyArray":
+                if getattr(typing, "GENERATING_DOCUMENTATION", False):
+
+                    class PRNGKeyArray:
+                        pass
+
+                    PRNGKeyArray.__module__ = "builtins"
+                    return PRNGKeyArray
+                else:
+                    # New-style `jax.random.key` have scalar shape and dtype `key<foo>`.
+                    # Old-style `jax.random.PRNGKey` have shape `(2,)` and dtype
+                    # `uint32`.
+                    import jax
+
+                    return Union[Key[jax.Array, ""], UInt32[jax.Array, "2"]]
+            elif item == "DTypeLike":
+                import jax.typing
+
+                return jax.typing.DTypeLike
+            elif item == "Scalar":
+                import jax
+
+                return Shaped[jax.Array, ""]
+            elif item == "ScalarLike":
+                import jax.typing
+
+                return Shaped[jax.typing.ArrayLike, ""]
+            elif item == "PyTree":
+                from ._pytree_type import PyTree
+
+                return PyTree
+            elif item == "PyTreeDef":
+                if hasattr(typing, "GENERATING_DOCUMENTATION"):
+                    # Most parts of the Equinox ecosystem have
+                    # `typing.GENERATING_DOCUMENTATION = True` when generating
+                    # documentation, to add whatever shims are necessary to get pretty
+                    # docs. E.g. to have type annotations appear as just `PyTree`, not
+                    # `jaxtyping.PyTree`.
+                    #
+                    # As jaxtyping actually wants things to appear as e.g.
+                    # `jaxtyping.PyTree`, rather than just `PyTree`, then it sets
+                    # `typing.GENERATING_DOCUMENTATION = False`, to disable these shims.
+                    #
+                    # Here we do only a `hasattr` check, as we want to get this version
+                    # of `PyTreeDef` in both the jaxtyping and the Equinox(/etc.) docs.
+
+                    class PyTreeDef:
+                        """Alias for `jax.tree_util.PyTreeDef`, which is the type of the
+                        return from `jax.tree_util.tree_structure(...)`.
+                        """
+
+                    if typing.GENERATING_DOCUMENTATION:
+                        # Equinox etc. docs get just `PyTreeDef`.
+                        # jaxtyping docs get `jaxtyping.PyTreeDef`.
+                        PyTreeDef.__module__ = "builtins"
+                    return PyTreeDef
+                else:
+                    import jax.tree_util
+
+                    return jax.tree_util.PyTreeDef
+            else:
+                raise AttributeError(f"module jaxtyping has no attribute {item!r}")
+
+
+check_equinox_version = True  # easy-to-replace line with copybara
+if check_equinox_version:
+    try:
+        eqx_version = importlib.metadata.version("equinox")
+    except importlib.metadata.PackageNotFoundError:
+        pass
     else:
-        from jax.tree_util import PyTreeDef as PyTreeDef
-
-    from ._pytree_type import PyTree as PyTree  # noqa: F401
-
-
-# Conveniences
-if typing.TYPE_CHECKING:
-    from jax.random import PRNGKeyArray as PRNGKeyArray
-
-    from ._indirection import Scalar as Scalar, ScalarLike as ScalarLike
-elif has_jax:
-    from ._array_types import Scalar, ScalarLike  # noqa: F401
-
-    if getattr(typing, "GENERATING_DOCUMENTATION", False):
-        # That is, we're generating some downstream documentation, not the jaxtyping
-        # documentation itself.
-        class PRNGKeyArray:
-            pass
-
-        PRNGKeyArray.__module__ = "builtins"
-    else:
-        from ._array_types import PRNGKeyArray
-
-del has_jax
+        major, minor, patch = eqx_version.split(".")
+        equinox_version = (int(major), int(minor), int(patch))
+        if equinox_version < (0, 11, 0):
+            warnings.warn(
+                "jaxtyping version >=0.2.23 should be used with Equinox version "
+                ">=0.11.1"
+            )
 
 
 __version__ = importlib.metadata.version("jaxtyping")

jaxtyping/_array_types.py

@@ -23,19 +23,18 @@ import re
 import sys
 import types
 import typing
+from dataclasses import dataclass
 from typing import Any, Literal, NoReturn, Optional, Union
 
 import numpy as np
 
-from ._decorator import storage
-
-
-try:
-    import jax
-except ImportError:
-    has_jax = False
-else:
-    has_jax = True
+from ._errors import AnnotationError
+from ._storage import (
+    get_shape_memo,
+    get_treeflatten_memo,
+    get_treepath_memo,
+    set_shape_memo,
+)
 
 
 _array_name_format = "dtype_and_shape"
@@ -52,7 +51,6 @@ def set_array_name_format(value):
 
 _any_dtype = object()
 
-
 _anonymous_dim = object()
 _anonymous_variadic_dim = object()
 
@@ -63,28 +61,30 @@ class _DimType(enum.Enum):
     symbolic = enum.auto()
 
 
+@dataclass(frozen=True)
 class _NamedDim:
-    def __init__(self, name, broadcastable):
-        self.name = name
-        self.broadcastable = broadcastable
+    name: str
+    broadcastable: bool
+    treepath: Any
 
 
+@dataclass(frozen=True)
 class _NamedVariadicDim:
-    def __init__(self, name, broadcastable):
-        self.name = name
-        self.broadcastable = broadcastable
+    name: str
+    broadcastable: bool
+    treepath: Any
 
 
+@dataclass(frozen=True)
 class _FixedDim:
-    def __init__(self, size, broadcastable):
-        self.size = size
-        self.broadcastable = broadcastable
+    size: str
+    broadcastable: bool
 
 
+@dataclass(frozen=True)
 class _SymbolicDim:
-    def __init__(self, expr, broadcastable):
-        self.expr = expr
-        self.broadcastable = broadcastable
+    elem: Any
+    broadcastable: bool
 
 
 _AbstractDimOrVariadicDim = Union[
@@ -100,9 +100,10 @@ _AbstractDim = Union[Literal[_anonymous_dim], _NamedDim, _FixedDim, _SymbolicDim
 
 def _check_dims(
     cls_dims: list[_AbstractDim],
-    obj_shape: tuple[int],
+    obj_shape: tuple[int, ...],
     single_memo: dict[str, int],
-) -> bool:
+    arg_memo: dict[str, Any],
+) -> str:
     assert len(cls_dims) == len(obj_shape)
     for cls_dim, obj_size in zip(cls_dims, obj_shape):
         if cls_dim is _anonymous_dim:
@@ -111,49 +112,58 @@ def _check_dims(
             pass
         elif type(cls_dim) is _FixedDim:
             if cls_dim.size != obj_size:
-                return False
+                return f"the dimension size {obj_size} does not equal {cls_dim.size} as expected by the type hint"  # noqa: E501
         elif type(cls_dim) is _SymbolicDim:
             try:
+                # Support f-string syntax.
+                # https://stackoverflow.com/a/53671539/22545467
+                elem = eval(f"f'{cls_dim.elem}'", arg_memo.copy())
                 # Make a copy to avoid `__builtins__` getting added as a key.
-                eval_size = eval(cls_dim.expr, single_memo.copy())
+                eval_size = eval(elem, single_memo.copy())
             except NameError as e:
-                raise NameError(
-                    f"Cannot process symbolic dimension '{cls_dim.expr}' as some "
-                    "dimension names have not been processed. In practice you should "
-                    "usually only use symbolic dimensions in annotations for return "
-                    "types, referring only to dimensions annotated for arguments."
+                raise AnnotationError(
+                    f"Cannot process symbolic axis '{cls_dim.elem}' as "
+                    "some axis names have not been processed. In practice you "
+                    "should usually only use symbolic axes in annotations "
+                    "for return types, referring only to axes annotated for "
+                    "arguments."
                 ) from e
             if eval_size != obj_size:
-                return False
+                return f"the dimension size {obj_size} does not equal the existing value of {cls_dim.elem}={eval_size}"  # noqa: E501
         else:
             assert type(cls_dim) is _NamedDim
+            if cls_dim.treepath:
+                name = get_treepath_memo() + cls_dim.name
+            else:
+                name = cls_dim.name
             try:
-                cls_size = single_memo[cls_dim.name]
+                cls_size = single_memo[name]
             except KeyError:
-                single_memo[cls_dim.name] = obj_size
+                single_memo[name] = obj_size
             else:
                 if cls_size != obj_size:
-                    return False
-    return True
-
-
-def _is_jax_extended_dtype(dtype: Any) -> bool:
-    if not has_jax:
-        return False
-    if hasattr(jax.dtypes, "extended"):  # jax>=0.4.14
-        return jax.numpy.issubdtype(dtype, jax.dtypes.extended)
-    else:  # jax<=0.4.13
-        return jax.core.is_opaque_dtype(dtype)
+                    return f"the size of dimension {cls_dim.name} is {obj_size} which does not equal the existing value of {cls_size}"  # noqa: E501
+    return ""
 
 
 class _MetaAbstractArray(type):
-    def __instancecheck__(cls, obj):
-        if not isinstance(obj, cls.array_type):
-            return False
+    _skip_instancecheck: bool = False
 
-        if _is_jax_extended_dtype(obj.dtype):
-            dtype = str(obj.dtype)
-        elif hasattr(obj.dtype, "type") and hasattr(obj.dtype.type, "__name__"):
+    def make_transparent(cls):
+        cls._skip_instancecheck = True
+
+    def __instancecheck__(cls, obj: Any) -> bool:
+        return cls.__instancecheck_str__(obj) == ""
+
+    def __instancecheck_str__(cls, obj: Any) -> str:
+        if cls._skip_instancecheck:
+            return ""
+        if not isinstance(obj, cls.array_type):
+            return f"this value is not an instance of the underlying array type {cls.array_type}"  # noqa: E501
+        if get_treeflatten_memo():
+            return ""
+
+        if hasattr(obj.dtype, "type") and hasattr(obj.dtype.type, "__name__"):
             # JAX, numpy
             dtype = obj.dtype.type.__name__
         elif hasattr(obj.dtype, "as_numpy_dtype"):
@@ -165,7 +175,7 @@ class _MetaAbstractArray(type):
             if len(repr_dtype) == 2 and repr_dtype[0] == "torch":
                 dtype = repr_dtype[1]
             else:
-                raise RuntimeError(
+                raise AnnotationError(
                     "Unrecognised array/tensor type to extract dtype from"
                 )
 
@@ -181,103 +191,133 @@ class _MetaAbstractArray(type):
                 if in_dtypes:
                     break
             if not in_dtypes:
-                return False
+                if len(cls.dtypes) == 1:
+                    return f"this array has dtype {dtype}, not {cls.dtypes[0]} as expected by the type hint"  # noqa: E501
+                else:
+                    return f"this array has dtype {dtype}, not any of {cls.dtypes} as expected by the type hint"  # noqa: E501
 
-        no_temp_memo = hasattr(storage, "memo_stack") and len(storage.memo_stack) != 0
-
-        if no_temp_memo:
-            single_memo, variadic_memo, variadic_broadcast_memo = storage.memo_stack[-1]
-            # Make a copy so we don't mutate the original memo during the shape check.
-            single_memo = single_memo.copy()
-            variadic_memo = variadic_memo.copy()
-            variadic_broadcast_memo = variadic_broadcast_memo.copy()
+        single_memo, variadic_memo, pytree_memo, arg_memo = get_shape_memo()
+        single_memo_bak = single_memo.copy()
+        variadic_memo_bak = variadic_memo.copy()
+        pytree_memo_bak = pytree_memo.copy()
+        arg_memo_bak = arg_memo.copy()
+        try:
+            check = cls._check_shape(obj, single_memo, variadic_memo, arg_memo)
+        except Exception:
+            set_shape_memo(
+                single_memo_bak, variadic_memo_bak, pytree_memo_bak, arg_memo_bak
+            )
+            raise
+        if check == "":
+            return check
         else:
-            # `isinstance` happening outside any @jaxtyped decorators, e.g. at the
-            # global scope. In this case just create a temporary memo, since we're not
-            # going to be comparing against any stored values anyway.
-            single_memo = {}
-            variadic_memo = {}
-            variadic_broadcast_memo = {}
-
-        if cls._check_shape(obj, single_memo, variadic_memo, variadic_broadcast_memo):
-            # We update the memo every time we successfully pass a shape check
-            if no_temp_memo:
-                storage.memo_stack[-1] = (
-                    single_memo,
-                    variadic_memo,
-                    variadic_broadcast_memo,
-                )
-            return True
-        else:
-            return False
+            set_shape_memo(
+                single_memo_bak, variadic_memo_bak, pytree_memo_bak, arg_memo_bak
+            )
+            return check
 
     def _check_shape(
         cls,
         obj,
         single_memo: dict[str, int],
-        variadic_memo: dict[str, tuple[int, ...]],
-        variadic_broadcast_memo: dict[str, list[tuple[int, ...]]],
-    ):
+        variadic_memo: dict[str, tuple[bool, tuple[int, ...]]],
+        arg_memo: dict[str, Any],
+    ) -> str:
         if cls.index_variadic is None:
             if obj.ndim != len(cls.dims):
-                return False
-            return _check_dims(cls.dims, obj.shape, single_memo)
+                return f"this array has {obj.ndim} dimensions, not the {len(cls.dims)} expected by the type hint"  # noqa: E501
+            return _check_dims(cls.dims, obj.shape, single_memo, arg_memo)
         else:
             if obj.ndim < len(cls.dims) - 1:
-                return False
+                return f"this array has {obj.ndim} dimensions, which is fewer than {len(cls.dims - 1)} that is the minimum expected by the type hint"  # noqa: E501
             i = cls.index_variadic
             j = -(len(cls.dims) - i - 1)
             if j == 0:
                 j = None
-            if not _check_dims(cls.dims[:i], obj.shape[:i], single_memo):
-                return False
-            if j is not None and not _check_dims(
-                cls.dims[j:], obj.shape[j:], single_memo
-            ):
-                return False
+            prefix_check = _check_dims(
+                cls.dims[:i], obj.shape[:i], single_memo, arg_memo
+            )
+            if prefix_check != "":
+                return prefix_check
+            if j is not None:
+                suffix_check = _check_dims(
+                    cls.dims[j:], obj.shape[j:], single_memo, arg_memo
+                )
+                if suffix_check != "":
+                    return suffix_check
             variadic_dim = cls.dims[i]
             if variadic_dim is _anonymous_variadic_dim:
-                return True
+                return ""
             else:
                 assert type(variadic_dim) is _NamedVariadicDim
-                variadic_name = variadic_dim.name
-                try:
-                    if variadic_dim.broadcastable:
-                        variadic_shapes = variadic_broadcast_memo[variadic_name]
-                    else:
-                        variadic_shape = variadic_memo[variadic_name]
-                except KeyError:
-                    if variadic_dim.broadcastable:
-                        variadic_broadcast_memo[variadic_name] = [obj.shape[i:j]]
-                    else:
-                        variadic_memo[variadic_name] = obj.shape[i:j]
-                    return True
+                if variadic_dim.treepath:
+                    name = get_treepath_memo() + variadic_dim.name
                 else:
-                    if variadic_dim.broadcastable:
-                        new_shape = obj.shape[i:j]
-                        for existing_shape in variadic_shapes:
-                            try:
-                                np.broadcast_shapes(new_shape, existing_shape)
-                            except ValueError:
-                                return False
-                        variadic_shapes.append(new_shape)
-                        return True
+                    name = variadic_dim.name
+                broadcastable = variadic_dim.broadcastable
+                try:
+                    prev_broadcastable, prev_shape = variadic_memo[name]
+                except KeyError:
+                    variadic_memo[name] = (broadcastable, obj.shape[i:j])
+                    return ""
+                else:
+                    new_shape = obj.shape[i:j]
+                    if prev_broadcastable:
+                        try:
+                            broadcast_shape = np.broadcast_shapes(new_shape, prev_shape)
+                        except ValueError:  # not broadcastable e.g. (3, 4) and (5,)
+                            return f"the shape of its variadic dimensions '*{variadic_dim.name}' is {new_shape}, which cannot be broadcast with the existing value of {prev_shape}"  # noqa: E501
+                        if not broadcastable and broadcast_shape != new_shape:
+                            return f"the shape of its variadic dimensions '*{variadic_dim.name}' is {new_shape}, which the existing value of {prev_shape} cannot be broadcast to"  # noqa: E501
+                        variadic_memo[name] = (broadcastable, broadcast_shape)
                     else:
-                        return variadic_shape == obj.shape[i:j]
+                        if broadcastable:
+                            try:
+                                broadcast_shape = np.broadcast_shapes(
+                                    new_shape, prev_shape
+                                )
+                            except ValueError:  # not broadcastable e.g. (3, 4) and (5,)
+                                return f"the shape of its variadic dimensions '*{variadic_dim.name}' is {new_shape}, which cannot be broadcast with the existing value of {prev_shape}"  # noqa: E501
+                            if broadcast_shape != prev_shape:
+                                return f"the shape of its variadic dimensions '*{variadic_dim.name}' is {new_shape}, which cannot be broadcast to the existing value of {prev_shape}"  # noqa: E501
+                        else:
+                            if new_shape != prev_shape:
+                                return f"the shape of its variadic dimensions '*{variadic_dim.name}' is {new_shape}, which does not equal the existing value of {prev_shape}"  # noqa: E501
+                    return ""
             assert False
 
 
 @ft.lru_cache(maxsize=None)
 def _make_metaclass(base_metaclass):
     class MetaAbstractArray(_MetaAbstractArray, base_metaclass):
-        pass
+        def _get_props(cls):
+            props_tuple = (
+                cls.index_variadic,
+                cls.dims,
+                cls.array_type,
+                cls.dtypes,
+                cls.dim_str,
+            )
+            return props_tuple
+
+        def __eq__(cls, other):
+            if type(cls) is not type(other):
+                return False
+
+            return cls._get_props() == other._get_props()
+
+        def __hash__(cls):
+            return hash(cls._get_props())
 
     return MetaAbstractArray
 
 
 def _check_scalar(dtype, dtypes, dims):
-    if len(dims) != 0:
-        return dims == (_anonymous_variadic_dim,)
+    for dim in dims:
+        if dim is not _anonymous_variadic_dim and not isinstance(
+            dim, _NamedVariadicDim
+        ):
+            return False
     return (_any_dtype is dtypes) or any(d.startswith(dtype) for d in dtypes)
 
 
@@ -298,14 +338,13 @@ class AbstractArray(metaclass=_MetaAbstractArray):
 
 _not_made = object()
 
-
 _union_types = [typing.Union]
 if sys.version_info >= (3, 10):
     _union_types.append(types.UnionType)
 
 
 @ft.lru_cache(maxsize=None)
-def _make_array(array_type, dim_str, dtypes, name):
+def _make_array_cached(array_type, dim_str, dtypes, name):
     if not isinstance(dim_str, str):
         raise ValueError(
             "Shape specification must be a string. Axes should be separated with "
@@ -317,28 +356,30 @@ def _make_array(array_type, dim_str, dtypes, name):
         if "," in elem and "(" not in elem:
             # Common mistake.
             # Disable in the case that there's brackets to allow for function calls,
-            # e.g. `min(foo,bar)`, in symbolic dimensions.
-            raise ValueError("Dimensions should be separated with spaces, not commas")
+            # e.g. `min(foo,bar)`, in symbolic axes.
+            raise ValueError("Axes should be separated with spaces, not commas")
         if elem.endswith("#"):
             raise ValueError(
-                "As of jaxtyping v0.1.0, broadcastable dimensions are now denoted "
+                "As of jaxtyping v0.1.0, broadcastable axes are now denoted "
                 "with a # at the start, rather than at the end"
             )
 
         if "..." in elem:
             if elem != "...":
                 raise ValueError(
-                    "Anonymous multiple dimension '...' must be used on its own; "
+                    "Anonymous multiple axes '...' must be used on its own; "
                     f"got {elem}"
                 )
             broadcastable = False
             variadic = True
             anonymous = True
+            treepath = False
             dim_type = _DimType.named
         else:
             broadcastable = False
             variadic = False
             anonymous = False
+            treepath = False
             while True:
                 if len(elem) == 0:
                     # This branch needed as just `_` is valid
@@ -356,7 +397,7 @@ def _make_array(array_type, dim_str, dtypes, name):
                     if variadic:
                         raise ValueError(
                             "Do not use * twice to denote accepting multiple "
-                            "dimensions, e.g. `**foo` is not allowed"
+                            "axes, e.g. `**foo` is not allowed"
                         )
                     variadic = True
                     elem = elem[1:]
@@ -368,6 +409,14 @@ def _make_array(array_type, dim_str, dtypes, name):
                         )
                     anonymous = True
                     elem = elem[1:]
+                elif first_char == "?":
+                    if treepath:
+                        raise ValueError(
+                            "Do not use ? twice to denote dependence on location "
+                            "within a PyTree, e.g. `??foo` is not allowed"
+                        )
+                    treepath = True
+                    elem = elem[1:]
                 # Allow e.g. `foo=4` as an alternate syntax for just `4`, so that one
                 # can write e.g. `Float[Array, "rows=3 cols=4"]`
                 elif elem.count("=") == 1:
@@ -387,28 +436,33 @@ def _make_array(array_type, dim_str, dtypes, name):
         if variadic:
             if index_variadic is not None:
                 raise ValueError(
-                    "Cannot use multiple-dimension specifiers (`*name` or `...`) "
-                    "more than once"
+                    "Cannot use variadic specifiers (`*name` or `...`) "
+                    "more than once."
                 )
             index_variadic = index
 
         if dim_type is _DimType.fixed:
             if variadic:
                 raise ValueError(
-                    "Cannot have a fixed axis bind to multiple dimensions, e.g. "
-                    "`*4` is not allowed"
+                    "Cannot have a fixed axis bind to multiple axes, e.g. "
+                    "`*4` is not allowed."
                 )
             if anonymous:
                 raise ValueError(
-                    "Cannot have a fixed axis be anonymous, e.g. `_4` is not " "allowed"
+                    "Cannot have a fixed axis be anonymous, e.g. `_4` is not allowed."
+                )
+            if treepath:
+                raise ValueError(
+                    "Cannot have a fixed axis have tree-path dependence, e.g. `?4` is "
+                    "not allowed."
                 )
             elem = _FixedDim(elem, broadcastable)
         elif dim_type is _DimType.named:
             if anonymous:
                 if broadcastable:
                     raise ValueError(
-                        "Cannot have a dimension be both anonymous and "
-                        "broadcastable, e.g. `#_` is not allowed"
+                        "Cannot have an axis be both anonymous and "
+                        "broadcastable, e.g. `#_` is not allowed."
                     )
                 if variadic:
                     elem = _anonymous_variadic_dim
@@ -416,22 +470,26 @@ def _make_array(array_type, dim_str, dtypes, name):
                     elem = _anonymous_dim
             else:
                 if variadic:
-                    elem = _NamedVariadicDim(elem, broadcastable)
+                    elem = _NamedVariadicDim(elem, broadcastable, treepath)
                 else:
-                    elem = _NamedDim(elem, broadcastable)
+                    elem = _NamedDim(elem, broadcastable, treepath)
         else:
             assert dim_type is _DimType.symbolic
             if anonymous:
                 raise ValueError(
-                    "Cannot have a symbolic dimension be anonymous, e.g. "
+                    "Cannot have a symbolic axis be anonymous, e.g. "
                     "`_foo+bar` is not allowed"
                 )
             if variadic:
                 raise ValueError(
-                    "Cannot have symbolic multiple-dimensions, e.g. "
+                    "Cannot have symbolic multiple-axes, e.g. "
                     "`*foo+bar` is not allowed"
                 )
-            elem = compile(elem, "<string>", "eval")
+            if treepath:
+                raise ValueError(
+                    "Cannot have a symbolic axis with tree-path dependence, e.g. "
+                    "`?foo+bar` is not allowed"
+                )
             elem = _SymbolicDim(elem, broadcastable)
         dims.append(elem)
     dims = tuple(dims)
@@ -459,6 +517,16 @@ def _make_array(array_type, dim_str, dtypes, name):
             return array_type
         else:
             return _not_made
+    elif array_type is np.bool_:
+        if _check_scalar("bool", dtypes, dims):
+            return array_type
+        else:
+            return _not_made
+    elif array_type is np.generic or array_type is np.number:
+        if _check_scalar("", dtypes, dims):
+            return array_type
+        else:
+            return _not_made
     if issubclass(array_type, AbstractArray):
         if dtypes is _any_dtype:
             dtypes = array_type.dtypes
@@ -475,7 +543,7 @@ def _make_array(array_type, dim_str, dtypes, name):
                 index_variadic = array_type.index_variadic + len(dims)
             else:
                 raise ValueError(
-                    "Cannot use multiple-dimension specifiers (`*name` or `...`) "
+                    "Cannot use variadic specifiers (`*name` or `...`) "
                     "in both the original array and the extended array"
                 )
         dims = dims + array_type.dims
@@ -491,28 +559,39 @@ def _make_array(array_type, dim_str, dtypes, name):
         name = type_str
     else:
         raise ValueError(f"array_name_format {_array_name_format} not recognised")
-    metaclass = _make_metaclass(type(array_type))
-    out = metaclass(
-        name,
-        (array_type, AbstractArray),
-        dict(
-            array_type=array_type,
-            dtypes=dtypes,
-            dims=dims,
-            index_variadic=index_variadic,
-            dim_str=dim_str,
-        ),
-    )
-    if getattr(typing, "GENERATING_DOCUMENTATION", False):
-        out.__module__ = "builtins"
-    else:
-        out.__module__ = "jaxtyping"
+
+    return (array_type, name, dtypes, dims, index_variadic, dim_str)
+
+
+def _make_array(*args, **kwargs):
+    out = _make_array_cached(*args, **kwargs)
+
+    if type(out) is tuple:
+        array_type, name, dtypes, dims, index_variadic, dim_str = out
+        metaclass = _make_metaclass(type(array_type))
+
+        out = metaclass(
+            name,
+            (array_type, AbstractArray),
+            dict(
+                array_type=array_type,
+                dtypes=dtypes,
+                dims=dims,
+                index_variadic=index_variadic,
+                dim_str=dim_str,
+            ),
+        )
+        if getattr(typing, "GENERATING_DOCUMENTATION", False):
+            out.__module__ = "builtins"
+        else:
+            out.__module__ = "jaxtyping"
+
     return out
 
 
 class _MetaAbstractDtype(type):
     def __instancecheck__(cls, obj: Any) -> NoReturn:
-        raise RuntimeError(
+        raise AnnotationError(
             f"Do not use `isinstance(x, jaxtyping.{cls.__name__})`. If you want to "
             "check just the dtype of an array, then use "
             f'`jaxtyping.{cls.__name__}[jnp.ndarray, "..."]`.'
@@ -521,10 +600,12 @@ class _MetaAbstractDtype(type):
     def __getitem__(cls, item: tuple[Any, str]):
         if not isinstance(item, tuple) or len(item) != 2:
             raise ValueError(
-                "As of jaxtyping v0.2.0, type annotations must now include an explicit "
-                "array type. For example `jaxtyping.Float32[jax.Array, 'foo bar']`."
+                "As of jaxtyping v0.2.0, type annotations must now include both an "
+                "array type and a shape. For example `Float[Array, 'foo bar']`.\n"
+                "Ellipsis can be used to accept any shape: `Float[Array, '...']`."
             )
         array_type, dim_str = item
+        dim_str = dim_str.strip()
         del item
         if typing.get_origin(array_type) in _union_types:
             out = [
@@ -590,12 +671,15 @@ class AbstractDtype(metaclass=_MetaAbstractDtype):
         cls.dtypes = dtypes
 
 
+_prng_key = "prng_key"
 _bool = "bool"
 _bool_ = "bool_"
+_uint4 = "uint4"
 _uint8 = "uint8"
 _uint16 = "uint16"
 _uint32 = "uint32"
 _uint64 = "uint64"
+_int4 = "int4"
 _int8 = "int8"
 _int16 = "int16"
 _int32 = "int32"
@@ -621,10 +705,12 @@ def _make_dtype(_dtypes, name):
     return _Cls
 
 
+UInt4 = _make_dtype(_uint4, "UInt4")
 UInt8 = _make_dtype(_uint8, "UInt8")
 UInt16 = _make_dtype(_uint16, "UInt16")
 UInt32 = _make_dtype(_uint32, "UInt32")
 UInt64 = _make_dtype(_uint64, "UInt64")
+Int4 = _make_dtype(_int4, "Int4")
 Int8 = _make_dtype(_int8, "Int8")
 Int16 = _make_dtype(_int16, "Int16")
 Int32 = _make_dtype(_int32, "Int32")
@@ -637,8 +723,8 @@ Complex64 = _make_dtype(_complex64, "Complex64")
 Complex128 = _make_dtype(_complex128, "Complex128")
 
 bools = [_bool, _bool_]
-uints = [_uint8, _uint16, _uint32, _uint64]
-ints = [_int8, _int16, _int32, _int64]
+uints = [_uint4, _uint8, _uint16, _uint32, _uint64]
+ints = [_int4, _int8, _int16, _int32, _int64]
 floats = [_bfloat16, _float16, _float32, _float64]
 complexes = [_complex64, _complex128]
 
@@ -652,17 +738,9 @@ Integer = _make_dtype(uints + ints, "Integer")
 Float = _make_dtype(floats, "Float")
 Complex = _make_dtype(complexes, "Complex")
 Inexact = _make_dtype(floats + complexes, "Inexact")
+Real = _make_dtype(floats + uints + ints, "Real")
 Num = _make_dtype(uints + ints + floats + complexes, "Num")
 
 Shaped = _make_dtype(_any_dtype, "Shaped")
 
-if has_jax:
-    if jax.config.jax_enable_custom_prng:
-        _key_regex = re.compile(r"^key<\w+>$")
-        Key = _make_dtype(_key_regex, "Key")
-        PRNGKeyArray = Key[jax.Array, ""]
-    else:
-        Key = UInt32
-        PRNGKeyArray = Key[jax.Array, "2"]
-    Scalar = Shaped[jax.Array, ""]
-    ScalarLike = Shaped[jax.typing.ArrayLike, ""]
+Key = _make_dtype(_prng_key, "Key")

jaxtyping/_config.py

@@ -0,0 +1,48 @@
+import os
+from typing import Union
+
+
+def _maybestr2bool(value: Union[bool, str], error: str) -> bool:
+    if isinstance(value, bool):
+        return value
+    elif isinstance(value, str):
+        if value.lower() in ("0", "false"):
+            return False
+        elif value.lower() in ("1", "true"):
+            return True
+        else:
+            raise ValueError(error)
+    else:
+        raise ValueError(error)
+
+
+class _JaxtypingConfig:
+    def __init__(self):
+        self.update("jaxtyping_disable", os.environ.get("JAXTYPING_DISABLE", "0"))
+        self.update(
+            "jaxtyping_remove_typechecker_stack",
+            os.environ.get("JAXTYPING_REMOVE_TYPECHECKER_STACK", "0"),
+        )
+
+    def update(self, item: str, value):
+        if item.lower() == "jaxtyping_disable":
+            msg = (
+                "Unrecognised value for `JAXTYPING_DISABLE`. Valid values are "
+                "`JAXTYPING_DISABLE=0` (the default) or `JAXTYPING_DISABLE=1` (to "
+                "disable runtime type checking)."
+            )
+            self.jaxtyping_disable = _maybestr2bool(value, msg)
+        elif item.lower() == "jaxtyping_remove_typechecker_stack":
+            msg = (
+                "Unrecognised value for `JAXTYPING_REMOVE_TYPECHECKER_STACK`. Valid "
+                "values are `JAXTYPING_REMOVE_TYPECHECKER_STACK=0` (the default) or "
+                "`JAXTYPING_REMOVE_TYPECHECKER_STACK=1` (to remove the stack frames "
+                "from the typechecker in `jaxtyped(typechecker=...)`, when it raises a "
+                "runtime type-checking error)."
+            )
+            self.jaxtyping_remove_typechecker_stack = _maybestr2bool(value, msg)
+        else:
+            raise ValueError(f"Unrecognised config value {item}")
+
+
+config = _JaxtypingConfig()

jaxtyping/_decorator.py

@@ -19,133 +19,782 @@
 
 import dataclasses
 import functools as ft
+import importlib.util
 import inspect
-import threading
-import types
-import weakref
+import itertools as it
+import sys
+import warnings
+from typing import Any, get_args, get_origin, get_type_hints, overload
+
+from jaxtyping import AbstractArray
+
+from ._config import config
+from ._errors import AnnotationError, TypeCheckError
+from ._storage import pop_shape_memo, push_shape_memo, shape_str
 
 
-try:
-    import jax._src.traceback_util as traceback_util
-except ImportError:
-    pass
-else:
-    traceback_util.register_exclusion(__file__)
+class _Sentinel:
+    def __repr__(self):
+        return "sentinel"
 
 
-storage = threading.local()
+_sentinel = _Sentinel()
+_tb_flag = True
 
 
-_jaxtyped_fns = weakref.WeakSet()
+@overload
+def jaxtyped(*, typechecker=_sentinel):
+    ...
 
 
-def jaxtyped(fn):
-    """Used in conjunction with a runtime type checker. Decorate a function with this to
-    have shapes checked for consistency across multiple arguments.
+@overload
+def jaxtyped(fn, *, typechecker=_sentinel):
+    ...
 
-    Note that `@jaxtyped` is applied above the type checker.
+
+def jaxtyped(fn=_sentinel, *, typechecker=_sentinel):
+    """Decorate a function with this to perform runtime type-checking of its arguments
+    and return value. Decorate a dataclass to perform type-checking of its attributes.
 
     !!! Example
 
         ```python
         # Import both the annotation and the `jaxtyped` decorator from `jaxtyping`
-        from jaxtyping import Array, Float32, jaxtyped
+        from jaxtyping import Array, Float, jaxtyped
 
         # Use your favourite typechecker: usually one of the two lines below.
         from typeguard import typechecked as typechecker
         from beartype import beartype as typechecker
 
-        # Write your function. @jaxtyped must be applied above @typechecker!
-        @jaxtyped
-        @typechecker
-        def batch_outer_product(x: Float32[Array, "b c1"],
-                                y: Float32[Array, "b c2"]
-                              ) -> Float32[Array, "b c1 c2"]:
+        # Type-check a function
+        @jaxtyped(typechecker=typechecker)
+        def batch_outer_product(x: Float[Array, "b c1"],
+                                y: Float[Array, "b c2"]
+                              ) -> Float[Array, "b c1 c2"]:
             return x[:, :, None] * y[:, None, :]
+
+        # Type-check a dataclass
+        from dataclasses import dataclass
+
+        @jaxtyped(typechecker=typechecker)
+        @dataclass
+        class MyDataclass:
+            x: int
+            y: Float[Array, "b c"]
         ```
 
-    **Notes for advanced users**
+    **Arguments:**
 
-    Put precisely, all `isinstance` shape checks are scoped to the thread-local dynamic
-    context of a `jaxtyped` call. A new dynamic context will allow different dimensions
-    sizes to be bound to the same name. After this new dynamic context is finished
-    then the old one is returned to.
+    - `fn`: The function or dataclass to decorate. In practice if you want to use
+        dataclasses with JAX, then
+        [`equinox.Module`](https://docs.kidger.site/equinox/api/module/module/) is our
+        recommended approach:
+        ```python
+        import equinox as eqx
 
-    For example, this means you could leave off the `@jaxtyped` decorator to enforce
-    that this function use the same axes sizes as the function it was called from.
+        @jaxtyped(typechecker=typechecker)
+        class MyModule(eqx.Module):
+            ...
+        ```
 
-    Likewise, this means you can use `isinstance` checks inside a function body
-    and have them contribute to the same collection of consistency checks performed
-    by a typechecker against its arguments. (Or even forgo a typechecker that analyses
-    arguments, and instead just do your own manual `isinstance` checks.)
+    - `typechecker`: Keyword-only argument: the runtime type-checker to use. This should
+        be a function decorator that will raise an exception if there is a type error,
+        e.g.
+        ```python
+        @typechecker
+        def f(x: int):
+            pass
 
-    Only `isinstance` checks that pass will contribute to the store of axis name-size
-    pairs; those that fail will not. As such it is safe to write e.g.
-    `assert not isinstance(x, Float32[Array, "foo"])`.
+        f("a string is not an integer")  # this line should raise an exception
+        ```
+        Common choices are `typechecker=beartype.beartype` or
+        `typechecker=typeguard.typechecked`. Can also be set as `typechecker=None` to
+        skip automatic runtime type-checking, but still support manual `isinstance`
+        checks inside the function body:
+        ```python
+        @jaxtyped(typechecker=None)
+        def f(x):
+            assert isinstance(x, Float[Array, "batch channel"])
+        ```
+
+    **Returns:**
+
+    If `fn` is a function (including a `staticmethod`, `classmethod`, or `property`),
+    then a wrapped function is returned.
+
+    If `fn` is a dataclass, then `fn` is returned directly, and additionally its
+    `__init__` method is wrapped and modified in-place.
+
+    !!! Info "Old syntax"
+
+        jaxtyping previously (before v0.2.24) recommended using this double-decorator
+        syntax:
+        ```python
+        @jaxtyped
+        @typechecker
+        def f(...): ...
+        ```
+        This is still supported, but will now raise a warning recommending the
+        `jaxtyped(typechecker=typechecker)` syntax discussed above. (Which will produce
+        easier-to-debug error messages: under the hood, the new syntax more carefully
+        manipulates the typechecker so as to determine where a type-check error arises.)
+
+    ??? Info "Notes for advanced users"
+
+        **Dynamic contexts:**
+
+        Put precisely, the axis names in e.g. `Float[Array, "batch channels"]` and the
+        structure names in e.g. `PyTree[int, "T"]` are all scoped to the thread-local
+        dynamic context of a `jaxtyped`-wrapped function. If from within that function
+        we then call another `jaxtyped`-wrapped function, then a new context is pushed
+        to the stack. The axis sizes and PyTree structures of this inner function will
+        then not be compared against the axis sizes and PyTree structures of the outer
+        function. After the inner function returns then this inner context is popped
+        from the stack, and the previous context is returned to.
+
+        **isinstance:**
+
+        Binding of a value against a name is done with an `isinstance` check, for
+        example `isinstance(jnp.zeros((3, 4)), Float[Array, "dim1 dim2"])` will bind
+        `dim1=3` and `dim2=4`. In practice these `isinstance` checks are usually done by
+        the run-time typechecker `typechecker` that is supplied as an argument.
+
+        This can also be done manually: add `isinstance` checks inside a function body
+        and they will contribute to the same collection of consistency checks as are
+        performed by the typechecker on the arguments and return values. (Or you can
+        forgo such a typechecker altogether -- i.e. `typechecker=None` -- and only do
+        your own manual `isinstance` checks.)
+
+        Only `isinstance` checks that pass will contribute to the store of values; those
+        that fail will not. As such it is safe to write e.g.
+        `assert not isinstance(x, Float32[Array, "foo"])`.
+
+        **Decoupling contexts from function calls:**
+
+        If you would like to call a new function *without* creating a new
+        dynamic context (and using the same set of axis and structure values), then
+        simply do not add a `jaxtyped` decorator to your inner function, whilst
+        continuing to perform type-checking in whatever way you prefer.
+
+        Conversely, if you would like a new dynamic context *without* calling a new
+        function, then in addition to the usage discussed above, `jaxtyped` also
+        supports being used as a context manager, by passing it the string `"context"`:
+        ```python
+        with jaxtyped("context"):
+            assert isinstance(x, Float[Array, "batch channel"])
+        ```
+        This is equivalent to placing this code inside a new function wrapped in
+        `jaxtyped(typechecker=None)`. Usage like this is very rare; it's mostly only
+        useful when working at the global scope.
     """
-    if type(fn) is types.FunctionType and fn in _jaxtyped_fns:
-        return fn
-    elif inspect.isclass(fn):  # allow decorators on class definitions
-        if dataclasses.is_dataclass(fn):
-            init = jaxtyped(fn.__init__)
-            fn.__init__ = init
-            return fn
-        else:
+
+    global _tb_flag
+    if _tb_flag and importlib.util.find_spec("jax._src.traceback_util") is not None:
+        import jax._src.traceback_util as traceback_util
+
+        traceback_util.register_exclusion(__file__)
+        _tb_flag = False
+
+    # First handle the `jaxtyped("context")` usage, which is a special case.
+    if fn == "context":
+        if typechecker is not _sentinel:
             raise ValueError(
-                "jaxtyped may only be added as a class decorator to dataclasses"
+                "Cannot use `jaxtyped` as a context with a typechecker. That is, "
+                "`with jaxtyped('context', typechecker=...):`. is not allowed. In this "
+                "case the type checker does not actually do anything, as there is no "
+                "function to type-check."
             )
+        return _JaxtypingContext()
+
+    # Now check that a typechecker has been explicitly declared. (Or explicitly declared
+    # as not being used, via `typechecker=None`.)
+    # This is needed just for backward compatibility: an undeclared typechecker
+    # corresponds to the old double-decorator syntax.
+    if typechecker is _sentinel:
+        # This branch will also catch the easy-to-make mistake of
+        # ```python
+        # @jaxtyped(typechecker)
+        # def foo(...):
+        # ```
+        # which is a bug as `typechecker` is interpreted as the function to decorate!
+        warnings.warn(
+            "As of jaxtyping version 0.2.24, jaxtyping now prefers the syntax\n"
+            "```\n"
+            "from jaxtyping import jaxtyped\n"
+            "# Use your favourite typechecker: usually one of the two lines below.\n"
+            "from typeguard import typechecked as typechecker\n"
+            "from beartype import beartype as typechecker\n"
+            "\n"
+            "@jaxtyped(typechecker=typechecker)\n"
+            "def foo(...):\n"
+            "```\n"
+            "and the old double-decorator syntax\n"
+            "```\n"
+            "@jaxtyped\n"
+            "@typechecker\n"
+            "def foo(...):\n"
+            "```\n"
+            "should no longer be used. (It will continue to work as it did before, but "
+            "the new approach will produce more readable error messages.)\n"
+            "In particular note that `typechecker` must be passed via keyword "
+            "argument; the following is not valid:\n"
+            "```\n"
+            "@jaxtyped(typechecker)\n"
+            "def foo(...):\n"
+            "```\n",
+            stacklevel=2,
+        )
+        typechecker = None
+
+    if fn is _sentinel:
+        return ft.partial(jaxtyped, typechecker=typechecker)
+    elif inspect.isclass(fn):
+        if dataclasses.is_dataclass(fn) and typechecker is not None:
+            # This does not check that the arguments passed to `__init__` match the
+            # type annotations. There may be a custom user `__init__`, or a
+            # dataclass-generated `__init__` used alongside
+            # `equinox.field(converter=...)`
+
+            init = fn.__init__
+
+            @ft.wraps(init)
+            def __init__(self, *args, **kwargs):
+                init(self, *args, **kwargs)
+                # `fn.__init__` is late-binding to the `__init__` function that
+                # we're in now. (Or to someone else's monkey-patch.) Either way,
+                # this checks that we're in the "top-level" `__init__`, and not one
+                # that is being called via `super()`. We don't want to trigger too
+                # early, before all fields have been assigned.
+                #
+                # We're not checking `if self.__class__ is fn` because Equinox
+                # replaces the with a defrozen version of itself during `__init__`,
+                # so the check wouldn't trigger.
+                #
+                # We're not doing this check by adding it to the end of the
+                # metaclass `__call__`, because Python doesn't allow you
+                # monkey-patch metaclasses.
+                if self.__class__.__init__ is fn.__init__:
+                    _check_dataclass_annotations(self, typechecker)
+
+            fn.__init__ = __init__
+        return fn
     # It'd be lovely if we could handle arbitrary descriptors, and not just the builtin
     # ones. Unfortunately that means returning a class instance with a __get__ method,
     # and that turns out to break loads of other things. See beartype issue #211 and
     # jaxtyping issue #71.
     elif isinstance(fn, classmethod):
-        return classmethod(jaxtyped(fn.__func__))
+        return classmethod(jaxtyped(fn.__func__, typechecker=typechecker))
     elif isinstance(fn, staticmethod):
-        return staticmethod(jaxtyped(fn.__func__))
+        return staticmethod(jaxtyped(fn.__func__, typechecker=typechecker))
     elif isinstance(fn, property):
         if fn.fget is None:
             fget = None
         else:
-            fget = jaxtyped(fn.fget)
+            fget = jaxtyped(fn.fget, typechecker=typechecker)
         if fn.fset is None:
             fset = None
         else:
-            fset = jaxtyped(fn.fset)
+            fset = jaxtyped(fn.fset, typechecker=typechecker)
         if fn.fdel is None:
             fdel = None
         else:
-            fdel = jaxtyped(fn.fdel)
+            fdel = jaxtyped(fn.fdel, typechecker=typechecker)
         return property(fget=fget, fset=fset, fdel=fdel)
     else:
+        if typechecker is None:
+            # Probably being used in the old style as
+            # ```
+            # @jaxtyped
+            # @typechecker
+            # def foo(x: int): ...
+            # ```
+            # in which case make a best-effort attempt to add shape information for any
+            # type errors.
 
-        @ft.wraps(fn)
-        def wrapped_fn(*args, **kwargs):
+            # we want to detect generators, and ignore return annotations on them,
+            # to avoid issues with O(n) typechecking trying to typecheck yielded values
+            wrp = fn
+            while hasattr(wrp, "__wrapped__"):
+                wrp = wrp.__wrapped__
+
+            if inspect.isgeneratorfunction(wrp) or inspect.isasyncgenfunction(wrp):
+                # recursively parse all the annotations, and mark all the jaxtyping
+                # annotations as not needing instance checks, while still being
+                # visible as original ones for the typechecker
+                def modify_annotation(ann):
+                    if inspect.isclass(ann) and issubclass(ann, AbstractArray):
+                        ann.make_transparent()
+
+                    for sub_ann in get_args(ann):
+                        modify_annotation(sub_ann)
+
+                # just to make sure: check that fn has valid return annotations
+                if hasattr(fn, "__annotations__") and "return" in fn.__annotations__:
+                    modify_annotation(fn.__annotations__["return"])
+
+            signature = inspect.signature(fn)
+
+            @ft.wraps(fn)
+            def wrapped_fn(*args, **kwargs):  # pyright: ignore
+                bound = signature.bind(*args, **kwargs)
+                memos = push_shape_memo(bound.arguments)
+                try:
+                    return fn(*args, **kwargs)
+                except Exception as e:
+                    # add_note api is support from python 3.11+
+                    if sys.version_info >= (3, 11) and _no_jaxtyping_note(e):
+                        shape_info = shape_str(memos)
+                        if shape_info != "":
+                            msg = (
+                                "The preceding error occurred within the scope of a "
+                                "`jaxtyping.jaxtyped` function, and may be due to a "
+                                "typecheck error. "
+                            )
+                            e.add_note(_jaxtyping_note_str(_spacer + msg + shape_info))
+                    raise
+                finally:
+                    pop_shape_memo()
+
+        else:
+            # New-style
+            # ```
+            # @jaxtyped(typechecker=typechecker)
+            # def foo(x: int): ...
+            # ```
+            # in which case we can do a better job reporting errors.
+
+            full_signature = inspect.signature(fn)
             try:
-                memo_stack = storage.memo_stack
-            except AttributeError:
-                memo_stack = storage.memo_stack = []
-            memo_stack.append(({}, {}, {}))
-            try:
-                return fn(*args, **kwargs)
-            finally:
-                memo_stack.pop()
+                destring_annotations = get_type_hints(fn, include_extras=True)
+            except NameError:
+                # Best-effort attempt to destringify annotations.
+                pass
+            else:
+                new_params = []
+                for p_name, p_value in full_signature.parameters.items():
+                    p_annotation = destring_annotations.get(p_name, p_value.annotation)
+                    p_value = p_value.replace(annotation=p_annotation)
+                    new_params.append(p_value)
+                return_annotation = destring_annotations.get(
+                    "return", full_signature.return_annotation
+                )
+                full_signature = full_signature.replace(
+                    parameters=new_params, return_annotation=return_annotation
+                )
+
+            param_signature = full_signature.replace(
+                return_annotation=inspect.Signature.empty
+            )
+            module = getattr(fn, "__module__", "generated")
+
+            full_fn, output_name = _make_fn_with_signature(
+                "check_return", full_signature, module, output=True
+            )
+            full_fn = typechecker(full_fn)
+
+            param_fn = _make_fn_with_signature(
+                "check_params", param_signature, module, output=False
+            )
+            param_fn = typechecker(param_fn)
+
+            @ft.wraps(fn)
+            def wrapped_fn(*args, **kwargs):
+                if config.jaxtyping_disable:
+                    return fn(*args, **kwargs)
+
+                # Raise bind-time errors before we do any shape analysis. (I.e. skip
+                # the pointless jaxtyping information for a non-typechecking failure.)
+                bound = param_signature.bind(*args, **kwargs)
+
+                memos = push_shape_memo(bound.arguments)
+                try:
+                    # First type-check just the parameters before the function is
+                    # called.
+                    try:
+                        param_fn(*args, **kwargs)
+                    except AnnotationError:
+                        raise
+                    except Exception as e:
+                        argmsg = _get_problem_arg(
+                            param_signature,
+                            args,
+                            kwargs,
+                            bound.arguments,
+                            module,
+                            typechecker,
+                        )
+                        try:
+                            name = fn.__name__
+                        except AttributeError:
+                            name = fn.__class__.__name__
+                        param_values = _pformat(bound.arguments, short_self=True)
+                        param_hints = _remove_typing(param_signature)
+                        msg = (
+                            "Type-check error whilst checking the parameters of "
+                            f"{name}.{argmsg}\n"
+                            "----------------------\n"
+                            f"Called with parameters: {param_values}\n"
+                            f"Parameter annotations: {param_hints}.\n"
+                            + shape_str(memos)
+                        )
+                        if config.jaxtyping_remove_typechecker_stack:
+                            raise TypeCheckError(msg) from None
+                        else:
+                            raise TypeCheckError(msg) from e
+
+                    # Actually call the function.
+                    out = fn(*args, **kwargs)
+
+                    if full_signature.return_annotation is not inspect.Signature.empty:
+                        # Now type-check the return value. We need to include the
+                        # parameters in the type-checking here in case there are any
+                        # type variables shared across the parameters and return.
+                        #
+                        # Incidentally this does mean that if `fn` mutates its arguments
+                        # so that they no longer satisfy their type annotations, this
+                        # will throw an error here. But that's like, super weird, so
+                        # don't do that. An error in that scenario is probably still
+                        # desirable.
+                        #
+                        # There is a small performance concern here when used in
+                        # non-jit'd contexts, like PyTorch, due to the duplicate
+                        # checking of the parameters. Unfortunately there doesn't seem
+                        # to be a way around that, so c'est la vie.
+                        kwargs[output_name] = out
+                        try:
+                            full_fn(*args, **kwargs)
+                        except AnnotationError:
+                            raise
+                        except Exception as e:
+                            try:
+                                name = fn.__name__
+                            except AttributeError:
+                                name = fn.__class__.__name__
+                            param_values = _pformat(bound.arguments, short_self=True)
+                            return_value = _pformat(out, short_self=False)
+                            param_hints = _remove_typing(param_signature)
+                            return_hint = _remove_typing(
+                                full_signature.return_annotation
+                            )
+                            if return_hint.startswith(
+                                "<class '"
+                            ) and return_hint.endswith("'>"):
+                                return_hint = return_hint[8:-2]
+                            msg = (
+                                "Type-check error whilst checking the return value "
+                                f"of {name}.\n"
+                                f"Actual value: {return_value}\n"
+                                f"Expected type: {return_hint}.\n"
+                                "----------------------\n"
+                                f"Called with parameters: {param_values}\n"
+                                f"Parameter annotations: {param_hints}.\n"
+                                + shape_str(memos)
+                            )
+                            if config.jaxtyping_remove_typechecker_stack:
+                                raise TypeCheckError(msg) from None
+                            else:
+                                raise TypeCheckError(msg) from e
+
+                    return out
+                finally:
+                    pop_shape_memo()
 
-        _jaxtyped_fns.add(wrapped_fn)
         return wrapped_fn
 
 
-def _jaxtyped_typechecker(typechecker):
-    # typechecker is expected to probably be either `typeguard.typechecked`, or
-    # `beartype.beartype`, or `None`.
+class _JaxtypingContext:
+    def __enter__(self):
+        push_shape_memo({})
 
-    if typechecker is None:
-        typechecker = lambda x: x
+    def __exit__(self, exc_type, exc_value, exc_tb):
+        pop_shape_memo()
 
-    def _wrapper(kls):
-        assert inspect.isclass(kls)
-        if dataclasses.is_dataclass(kls):
-            init = jaxtyped(typechecker(kls.__init__))
-            kls.__init__ = init
-        return kls
 
-    return _wrapper
+def _check_dataclass_annotations(self, typechecker):
+    """Creates and calls a function that checks the attributes of `self`
+
+    `self` should be a dataclass instance. `typechecker` should be e.g.
+    `beartype.beartype` or `typeguard.typechecked`.
+    """
+    parameters = [inspect.Parameter("self", inspect.Parameter.POSITIONAL_OR_KEYWORD)]
+    values = {}
+    for field in dataclasses.fields(self):
+        annotation = field.type
+        if isinstance(annotation, str):
+            # Don't check stringified annotations. These are basically impossible to
+            # resolve correctly, so just skip them.
+            continue
+        if get_origin(annotation) is type:
+            args = get_args(annotation)
+            if len(args) == 1 and isinstance(args[0], str):
+                # We also special-case this one kind of partially-stringified type
+                # annotation, so as to support Equinox <v0.11.1.
+                # This was fixed in Equinox in
+                # https://github.com/patrick-kidger/equinox/pull/543
+                continue
+        try:
+            value = getattr(self, field.name)  # noqa: F841
+        except AttributeError:
+            continue  # allow uninitialised fields, which are allowed on dataclasses
+
+        parameters.append(
+            inspect.Parameter(
+                field.name,
+                inspect.Parameter.POSITIONAL_OR_KEYWORD,
+                annotation=field.type,
+            )
+        )
+        values[field.name] = value
+
+    signature = inspect.Signature(parameters)
+    module = self.__class__.__module__
+    f = _make_fn_with_signature(
+        self.__class__.__name__, signature, module, output=False
+    )
+    f = jaxtyped(f, typechecker=typechecker)
+    f(self, **values)
+
+
+def _make_fn_with_signature(
+    name: str, signature: inspect.Signature, module: str, output: bool
+):
+    """Dynamically creates a function `fn` with name `name` and signature `signature`.
+
+    If `output=True` then `fn` will consume an additional keyword-only argument (in
+    addition to the provided signature), and will directly return this argument. In this
+    case the returned value from `_make_fn_with_signature` is a 2-tuple of `(fn, name)`,
+    where `fn` is the generated function, and `name` is the name of this extra argument.
+
+    If `output=False` then `fn` will just have a single `pass` statement, and the
+    returned value from `_make_fn_with_signature` will just be `fn`.
+
+    ---
+
+    Note that this function operates by dynamically creating and eval'ing a string, not
+    simply by assigning `__signature__` and `__annotations__`. The latter is enough for
+    typeguard (at least v2), but does not work with beartype (at least v16).
+    """
+    pos = []
+    pos_or_key = []
+    varpos = []
+    key = []
+    varkey = []
+    for p in signature.parameters.values():
+        if p.kind == inspect.Parameter.POSITIONAL_ONLY:
+            pos.append(p)
+        elif p.kind == inspect.Parameter.POSITIONAL_OR_KEYWORD:
+            pos_or_key.append(p)
+        elif p.kind == inspect.Parameter.VAR_POSITIONAL:
+            varpos.append(p)
+        elif p.kind == inspect.Parameter.KEYWORD_ONLY:
+            key.append(p)
+        elif p.kind == inspect.Parameter.VAR_KEYWORD:
+            varkey.append(p)
+        else:
+            assert False
+
+    param_names = frozenset(signature.parameters.keys())
+    if output:
+        output_name = _gensym(param_names, prefix="ret")
+        outstr = "return " + output_name
+        param_names = param_names | frozenset({output_name})
+        key.append(inspect.Parameter(output_name, kind=inspect.Parameter.KEYWORD_ONLY))
+    else:
+        outstr = "pass"
+
+    scope = {name: None}
+    name_to_annotation = {}
+    name_to_default = {}
+    param_triples = (
+        (p.name, p.annotation, p.default) for p in signature.parameters.values()
+    )
+    if output:
+        triples = it.chain(
+            param_triples,
+            [
+                ("return", signature.return_annotation, inspect.Signature.empty),
+                (output_name, Any, inspect.Signature.empty),
+            ],
+        )
+    else:
+        triples = it.chain(
+            param_triples,
+            [("return", signature.return_annotation, inspect.Signature.empty)],
+        )
+    for p_name, p_annotation, p_default in triples:
+        annotation_name = _gensym(frozenset(scope.keys()) | param_names, prefix="T")
+        name_to_annotation[p_name] = annotation_name
+        if p_annotation is inspect.Signature.empty or isinstance(p_annotation, str):
+            # If we have a stringified annotation here it's because the get_type_hints
+            # lookup above failed. Typically this occurs when using a local variable as
+            # the annotation. In this case we really have no idea what the annotation
+            # refers to, so just set it to Any.
+            # This does mean that we don't handle partially-stringified local
+            # annotations, e.g. `type["Foo"]` for some local type `Foo`. Those will
+            # probably just error out. Nothing better we can do about that
+            # unfortunately.
+            scope[annotation_name] = Any
+        else:
+            scope[annotation_name] = p_annotation
+        default_name = _gensym(frozenset(scope.keys()) | param_names, prefix="default")
+        name_to_default[p_name] = default_name
+        scope[default_name] = p_default
+
+    argstr_pieces = []
+    if len(pos) > 0:
+        for p in pos:
+            argstr_pieces.append(_make_argpiece(p, name_to_annotation, name_to_default))
+        argstr_pieces.append("/")
+    if len(pos_or_key) > 0:
+        for p in pos_or_key:
+            argstr_pieces.append(_make_argpiece(p, name_to_annotation, name_to_default))
+    if len(varpos) == 1:
+        [p] = varpos
+        argstr_pieces.append(
+            "*" + _make_argpiece(p, name_to_annotation, name_to_default)
+        )
+    else:
+        assert len(varpos) == 0
+        if len(key) > 0:
+            argstr_pieces.append("*")
+    if len(key) > 0:
+        for p in key:
+            argstr_pieces.append(_make_argpiece(p, name_to_annotation, name_to_default))
+    if len(varkey) == 1:
+        [p] = varkey
+        argstr_pieces.append(
+            "**" + _make_argpiece(p, name_to_annotation, name_to_default)
+        )
+    else:
+        assert len(varkey) == 0
+    argstr = ", ".join(argstr_pieces)
+
+    if signature.return_annotation is inspect.Signature.empty:
+        retstr = ""
+    else:
+        retstr = f"-> {name_to_annotation['return']}"
+
+    fnstr = f"def {name}({argstr}){retstr}:\n    {outstr}"
+    exec(fnstr, scope)
+    fn = scope[name]
+    fn.__module__ = module
+    assert fn is not None
+    if output:
+        return fn, output_name
+    else:
+        return fn
+
+
+def _gensym(names: frozenset[str], prefix: str) -> str:
+    assert prefix.isidentifier()
+    output_index = 0
+    output_name = prefix + str(output_index)
+    while output_name in names:
+        output_index += 1
+        output_name = prefix + str(output_index)
+    assert output_name.isidentifier()
+    return output_name
+
+
+def _make_argpiece(p, name_to_annotation, name_to_default):
+    if p.default is inspect.Signature.empty:
+        return f"{p.name}: {name_to_annotation[p.name]}"
+    else:
+        return f"{p.name}: {name_to_annotation[p.name]} = {name_to_default[p.name]}"
+
+
+def _get_problem_arg(
+    param_signature: inspect.Signature, args, kwargs, arguments, module, typechecker
+) -> str:
+    """Determines which argument was likely to be the problematic one responsible for
+    raising a type-check error.
+    """
+    # No performance concerns, as this is only used when we're about to raise an error
+    # anyway.
+    for keep_name in param_signature.parameters.keys():
+        new_parameters = []
+        keep_annotation = sentinel = object()
+        for p_name, p in param_signature.parameters.items():
+            if p_name == keep_name:
+                new_parameters.append(
+                    inspect.Parameter(p.name, p.kind, annotation=p.annotation)
+                )
+                assert keep_annotation is sentinel
+                keep_annotation = _remove_typing(p.annotation)
+            else:
+                new_parameters.append(inspect.Parameter(p.name, p.kind))
+        assert keep_annotation is not sentinel
+        new_signature = inspect.Signature(new_parameters)
+        fn = _make_fn_with_signature(
+            "check_single_arg", new_signature, module, output=False
+        )
+        fn = typechecker(fn)  # but no `jaxtyped`; keep the same environment.
+        try:
+            fn(*args, **kwargs)
+        except Exception:
+            keep_value = _pformat(arguments[keep_name], short_self=False)
+            return (
+                f"\nThe problem arose whilst typechecking parameter '{keep_name}'.\n"
+                f"Actual value: {keep_value}\n"
+                f"Expected type: {keep_annotation}."
+            )
+    else:
+        # Could not localise the problem to a single argument -- probably due to
+        # e.g. a mismatched typevar, which each individual argument is okay with.
+        return ""
+
+
+def _remove_typing(x):
+    x = str(x)
+    x = x.replace(" jaxtyping.", " ")
+    x = x.replace("[jaxtyping.", "[")
+    x = x.replace("'jaxtyping.", "'")
+    x = x.replace(" typing.", " ")
+    x = x.replace("[typing.", "[")
+    x = x.replace("'typing.", "'")
+    return x
+
+
+def _pformat(x, short_self: bool):
+    # No performance concerns from delayed imports -- this is only used when we're about
+    # to raise an error anyway.
+    try:
+        # TODO(kidger): this is pretty ugly. We have a circular dependency
+        # equinox->jaxtyping->equinox. We could consider moving all the pretty-printing
+        # code from equinox into jaxtyping maybe? Or into some shared dependency?
+        import equinox as eqx
+
+        pformat = eqx.tree_pformat
+        if short_self:
+            try:
+                self = x["self"]
+            except KeyError:
+                pass
+            else:
+                is_self = lambda y: y is self
+                pformat = ft.partial(pformat, truncate_leaf=is_self)
+    except Exception:
+        import pprint
+
+        pformat = ft.partial(pprint.pformat, indent=2, compact=True)
+    return pformat(x)
+
+
+class _jaxtyping_note_str(str):
+    """Used with `_no_jaxtyping_note` to flag that a note came from jaxtyping."""
+
+
+def _no_jaxtyping_note(e: Exception) -> bool:
+    """Checks if any of the exception's notes are from jaxtyping."""
+    try:
+        notes = e.__notes__
+    except AttributeError:
+        return True
+    else:
+        for note in notes:
+            if isinstance(note, _jaxtyping_note_str):
+                return False
+        return True
+
+
+_spacer = "--------------------\n"

jaxtyping/_errors.py

@@ -0,0 +1,12 @@
+class TypeCheckError(TypeError):
+    pass
+
+
+# Not inheriting from TypeError as that gets caught and re-reraised as just a TypeError
+# when using typeguard<3.
+class AnnotationError(Exception):
+    pass
+
+
+TypeCheckError.__module__ = "jaxtyping"
+AnnotationError.__module__ = "jaxtyping"

jaxtyping/_import_hook.py

@@ -76,32 +76,72 @@ def _optimized_cache_from_source(typechecker_hash, /, path, debug_override=None)
     # Version 5: Added support for string-based `typechecker` argument.
     # Version 6: optimization tag now depends on `typechecker` argument, so that
     #    changing the typechecker will hit a different cache.
+    # Version 7: Using the same md5 hash of the `typechecker` argument
+    #    for importlib and decorator lookup.
+    # Version 8: Now using new-style `jaxtyped(typechecker=...)` rather than old-style
+    #    double-decorators.
     return cache_from_source(
-        path, debug_override, optimization=f"jaxtyping6{typechecker_hash}"
+        path, debug_override, optimization=f"jaxtyping8{typechecker_hash}"
     )
 
 
-def _dot_lookup(*elements):
-    out = ast.Name(id=elements[0], ctx=ast.Load())
-    for element in elements[1:]:
-        out = ast.Attribute(out, element, ctx=ast.Load())
-    return out
+class Typechecker:
+    lookup = {}
+
+    def __init__(self, typechecker):
+        self.ast = None
+
+        if isinstance(typechecker, str):
+            # If the typechecker is a string, then we parse it
+            string_to_eval = (
+                "def f(x, *args, **kwargs):\n"
+                + f"  import {typechecker.split('.', 1)[0]}\n"
+                + f"  return {typechecker}(x, *args, **kwargs)"
+            )
+
+            # md5 hashing instead of __hash__
+            # because __hash__ is different for each Python session
+            self.hash = hashlib.md5(typechecker.encode("utf-8")).hexdigest()
+
+            vars = {}
+            exec(string_to_eval, {}, vars)
+            Typechecker.lookup[self.hash] = vars["f"]
+
+        elif typechecker is None:
+            # If it is None, ignore it silently (use dummy decorator)
+            self.hash = "0"
+            Typechecker.lookup[self.hash] = lambda x, *_, **__: x
+        else:
+            # Passed typechecker is invalid
+            raise TypeError(
+                "Jaxtyping typechecker has to be either a string or a None."
+            )
+
+    def get_hash(self):
+        return self.hash
+
+    def get_ast(self):
+        # we compile AST only if we missed importlib cache
+        if self.ast is None:
+            self.ast = (
+                ast.parse(
+                    f"@jaxtyping.jaxtyped(typechecker=jaxtyping._import_hook.Typechecker.lookup['{self.hash}'])\n"
+                    "def _():\n    ..."
+                )
+                .body[0]
+                .decorator_list[0]
+            )
+
+        return self.ast
 
 
-def _str_lookup(string):
-    module = ast.parse(string)
-    (expr,) = module.body
-    return expr.value
-
-
-class _JaxtypingTransformer(ast.NodeVisitor):
-    def __init__(self, *, typechecker) -> None:
+class JaxtypingTransformer(ast.NodeVisitor):
+    def __init__(self, *, typechecker: Typechecker) -> None:
         self._parents: list[ast.AST] = []
         self._typechecker = typechecker
 
     def visit_Module(self, node: ast.Module):
-        # Insert "import typeguard; import jaxtping" after any "from __future__ ..."
-        # imports
+        # Insert "import jaxtyping" after any "from __future__ ..." imports
         for i, child in enumerate(node.body):
             if isinstance(child, ast.ImportFrom) and child.module == "__future__":
                 continue
@@ -109,11 +149,6 @@ class _JaxtypingTransformer(ast.NodeVisitor):
                 continue  # module docstring
             else:
                 node.body.insert(i, ast.Import(names=[ast.alias("jaxtyping", None)]))
-                if self._typechecker is not None:
-                    typechecker_module, _ = self._typechecker.split(".", 1)
-                    node.body.insert(
-                        i, ast.Import(names=[ast.alias(typechecker_module, None)])
-                    )
                 break
 
         self._parents.append(node)
@@ -122,12 +157,9 @@ class _JaxtypingTransformer(ast.NodeVisitor):
         return node
 
     def visit_ClassDef(self, node: ast.ClassDef):
-        func = _dot_lookup("jaxtyping", "_decorator", "_jaxtyped_typechecker")
-        if self._typechecker is None:
-            args = [ast.Constant(None)]
-        else:
-            args = [_str_lookup(self._typechecker)]
-        node.decorator_list.insert(0, ast.Call(func, args, keywords=[]))
+        # Place at the start of the decorator list, so that `@dataclass` decorators get
+        # called first.
+        node.decorator_list.insert(0, self._typechecker.get_ast())
         self._parents.append(node)
         self.generic_visit(node)
         self._parents.pop()
@@ -137,8 +169,13 @@ class _JaxtypingTransformer(ast.NodeVisitor):
         has_annotated_args = any(arg for arg in node.args.args if arg.annotation)
         has_annotated_return = bool(node.returns)
         if has_annotated_args or has_annotated_return:
-            # Place at the end of the decorator list, as otherwise we wrap e.g.
-            # `jax.custom_{jvp,vjp}` and lose the ability to `defjvp` etc.
+            # Place at the end of the decorator list, because:
+            # - as otherwise we wrap e.g. `jax.custom_{jvp,vjp}` and lose the ability
+            #     to `defjvp` etc.
+            # - decorators frequently remove annotations from functions, and we'd like
+            #     to use those annotations.
+            # - typeguard in particular wants to be at the end of the decorator list, as
+            #     it works by recompling the wrapped function.
             #
             # Note that the counter-argument here is that we'd like to place this
             # at the start of the decorator list, in case a typechecking annotation
@@ -146,16 +183,8 @@ class _JaxtypingTransformer(ast.NodeVisitor):
             # case we're just going to have to need to ask the user to remove their
             # typechecking annotation (and let this decorator do it instead).
             # It's more important we be compatible with normal JAX code.
-            #
-            #
-            # FWIW, typeguard also wants to be at the end of the decorator list, as it
-            # works by recompiling the wrapped function.
-            node.decorator_list.append(_dot_lookup("jaxtyping", "jaxtyped"))
-            if self._typechecker is not None:
-                # Place at the end of the decorator list, as decorators
-                # frequently remove annotations from functions and we'd like to
-                # use those annotations.
-                node.decorator_list.append(_str_lookup(self._typechecker))
+            node.decorator_list.append(self._typechecker.get_ast())
+
         self._parents.append(node)
         self.generic_visit(node)
         self._parents.pop()
@@ -163,12 +192,9 @@ class _JaxtypingTransformer(ast.NodeVisitor):
 
 
 class _JaxtypingLoader(SourceFileLoader):
-    def __init__(self, *args, typechecker, **kwargs):
+    def __init__(self, *args, typechecker: Typechecker, **kwargs):
         super().__init__(*args, **kwargs)
         self._typechecker = typechecker
-        self._typechecker_hash = hashlib.md5(
-            self._typechecker.encode("utf-8")
-        ).hexdigest()
 
     def source_to_code(self, data, path, *, _optimize=-1):
         source = decode_source(data)
@@ -181,7 +207,7 @@ class _JaxtypingLoader(SourceFileLoader):
             dont_inherit=True,
             optimize=_optimize,
         )
-        tree = _JaxtypingTransformer(typechecker=self._typechecker).visit(tree)
+        tree = JaxtypingTransformer(typechecker=self._typechecker).visit(tree)
         ast.fix_missing_locations(tree)
         return _call_with_frames_removed(
             compile, tree, path, "exec", dont_inherit=True, optimize=_optimize
@@ -192,7 +218,7 @@ class _JaxtypingLoader(SourceFileLoader):
         # patch safe
         with patch(
             "importlib._bootstrap_external.cache_from_source",
-            ft.partial(_optimized_cache_from_source, self._typechecker_hash),
+            ft.partial(_optimized_cache_from_source, self._typechecker.get_hash()),
         ):
             return super().exec_module(module)
 
@@ -204,7 +230,7 @@ class _JaxtypingFinder(MetaPathFinder):
     Should not be used directly, but rather via `install_import_hook`.
     """
 
-    def __init__(self, modules, original_pathfinder, typechecker):
+    def __init__(self, modules, original_pathfinder, typechecker: Typechecker):
         self.modules = modules
         self._original_pathfinder = original_pathfinder
         self._typechecker = typechecker
@@ -255,7 +281,8 @@ class ImportHookManager:
 # Deliberately no default for `typechecker` so that folks must opt-in to not having
 # a typechecker.
 def install_import_hook(modules: Union[str, Sequence[str]], typechecker: Optional[str]):
-    """Automatically apply `@jaxtyped`, and optionally a type checker, as decorators.
+    """Automatically apply the `@jaxtyped(typechecker=typechecker)` decorator to every
+    function and dataclass over a whole codebase.
 
     !!! Tip "Usage"
 
@@ -263,18 +290,19 @@ def install_import_hook(modules: Union[str, Sequence[str]], typechecker: Optiona
         from jaxtyping import install_import_hook
         # Plus any one of the following:
 
-        # decorate @jaxtyped and @typeguard.typechecked
+        # decorate `@jaxtyped(typechecker=typeguard.typechecked)`
         with install_import_hook("foo", "typeguard.typechecked"):
             import foo          # Any module imported inside this `with` block, whose
             import foo.bar      # name begins with the specified string, will
             import foo.bar.qux  # automatically have both `@jaxtyped` and the specified
-                                # typechecker applied to all of their functions.
+                                # typechecker applied to all of their functions and
+                                # dataclasses.
 
-        # decorate @jaxtyped and @beartype.beartype
+        # decorate `@jaxtyped(typechecker=beartype.beartype)`
         with install_import_hook("foo", "beartype.beartype"):
             ...
 
-        # decorate only @jaxtyped (if you want that for some reason)
+        # decorate only `@jaxtyped` (if you want that for some reason)
         with install_import_hook("foo", None):
             ...
         ```
@@ -291,22 +319,9 @@ def install_import_hook(modules: Union[str, Sequence[str]], typechecker: Optiona
         install_import_hook(["foo", "bar.baz"], ...)
         ```
 
-    The import hook will automatically decorate all functions, and the `__init__` method
-    of dataclasses.
-
-    If the function already has any decorators on it, then both the `@jaxtyped` and the
-    typechecker decorators will get added at the bottom of the decorator list, e.g.
-    ```python
-    @some_other_decorator
-    @jaxtyped
-    @beartype.beartype
-    def foo(...): ...
-    ```
-
     **Arguments:**:
 
-    - `modules`: the names of the modules in which to automatically apply `@jaxtyped`
-        and `@typechecked`.
+    - `modules`: the names of the modules in which to automatically apply `@jaxtyped`.
     - `typechecker`: the module and function of the typechecker you want to use, as a
         string. For example `typechecker="typeguard.typechecked"`, or
         `typechecker="beartype.beartype"`. You may pass `typechecker=None` if you do not
@@ -316,7 +331,7 @@ def install_import_hook(modules: Union[str, Sequence[str]], typechecker: Optiona
 
     A context manager that uninstalls the hook on exit, or when you call `.uninstall()`.
 
-    ??? Example "Example: end-user script"
+    !!! Example "Example: end-user script"
 
         ```python
         ### entry_point.py
@@ -331,7 +346,7 @@ def install_import_hook(modules: Union[str, Sequence[str]], typechecker: Optiona
             ...
         ```
 
-    ??? Example "Example: writing a library"
+    !!! Example "Example: writing a library"
 
         ```python
         ### __init__.py
@@ -343,29 +358,28 @@ def install_import_hook(modules: Union[str, Sequence[str]], typechecker: Optiona
                                                  # so will be hook'd.
         ```
 
-    ??? info "Pytest hook"
+    !!! warning
 
-        The import hook can be installed at test-time only, as a pytest hook. From the
-        command line the syntax is:
+        If a function already has any decorators on it, then `@jaxtyped` will get added
+        at the bottom of the decorator list, e.g.
+        ```python
+        @some_other_decorator
+        @jaxtyped(typechecker=beartype.beartype)
+        def foo(...): ...
         ```
-        pytest --jaxtyping-packages=foo,bar.baz,beartype.beartype
-        ```
-        or in `pyproject.toml`:
-        ```toml
-        [tool.pytest.ini_options]
-        addopts = "--jaxtyping-packages=foo,bar.baz,beartype.beartype"
-        ```
-        or in `pytest.ini`:
-        ```ini
-        [pytest]
-        addopts = --jaxtyping-packages=foo,bar.baz,beartype.beartype
-        ```
-        This example will apply the import hook to all modules whose names start with
-        either `foo` or `bar.baz`. The typechecker used in this example is
-        `beartype.beartype`.
+        This is to support the common case in which
+        `some_other_decorator = jax.custom_jvp` etc.
 
-        (This is the author's preferred approach to performing runtime type-checking
-        with jaxtyping!)
+        If a class already has any decorators in it, then `@jaxtyped` will get added to
+        the top of the decorator list, e.g.
+        ```python
+        @jaxtyped(typechecker=beartype.beartype)
+        @some_other_decorator
+        class A:
+            ...
+        ```
+        This is to support the common case in which
+        `some_other_decorator = dataclasses.dataclass`.
     """  # noqa: E501
 
     if isinstance(modules, str):
@@ -385,6 +399,7 @@ def install_import_hook(modules: Union[str, Sequence[str]], typechecker: Optiona
     else:
         raise RuntimeError("Cannot find a PathFinder in sys.meta_path")
 
-    hook = _JaxtypingFinder(modules, finder, typechecker)
+    wrapped_typechecker = Typechecker(typechecker)
+    hook = _JaxtypingFinder(modules, finder, wrapped_typechecker)
     sys.meta_path.insert(0, hook)
     return ImportHookManager(hook)

jaxtyping/_indirection.py

@@ -1,3 +1,22 @@
+# Copyright (c) 2022 Google LLC
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy of
+# this software and associated documentation files (the "Software"), to deal in
+# the Software without restriction, including without limitation the rights to
+# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
+# the Software, and to permit persons to whom the Software is furnished to do so,
+# subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
+# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
+# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
 # Note that `from typing import Annotated; Bool = Annotated`
 # does not work with static type checkers. `Annotated` is a typeform rather
 # than a type, meaning it cannot be assigned.
@@ -13,6 +32,7 @@ from typing import (
     Annotated as Float64,  # noqa: F401
     Annotated as Inexact,  # noqa: F401
     Annotated as Int,  # noqa: F401
+    Annotated as Int4,  # noqa: F401
     Annotated as Int8,  # noqa: F401
     Annotated as Int16,  # noqa: F401
     Annotated as Int32,  # noqa: F401
@@ -20,13 +40,23 @@ from typing import (
     Annotated as Integer,  # noqa: F401
     Annotated as Key,  # noqa: F401
     Annotated as Num,  # noqa: F401
+    Annotated as Real,  # noqa: F401
     Annotated as Shaped,  # noqa: F401
     Annotated as UInt,  # noqa: F401
+    Annotated as UInt4,  # noqa: F401
     Annotated as UInt8,  # noqa: F401
     Annotated as UInt16,  # noqa: F401
     Annotated as UInt32,  # noqa: F401
     Annotated as UInt64,  # noqa: F401
+    TYPE_CHECKING,
 )
 
-from jax import Array as Scalar  # noqa: F401
+
+if not TYPE_CHECKING:
+    assert False
+
+from jax import (
+    Array as PRNGKeyArray,  # noqa: F401
+    Array as Scalar,  # noqa: F401
+)
 from jax.typing import ArrayLike as ScalarLike  # noqa: F401

jaxtyping/_ipython_extension.py

@@ -0,0 +1,56 @@
+# Copyright (c) 2022 Google LLC
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy of
+# this software and associated documentation files (the "Software"), to deal in
+# the Software without restriction, including without limitation the rights to
+# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
+# the Software, and to permit persons to whom the Software is furnished to do so,
+# subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
+# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
+# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+from ._import_hook import JaxtypingTransformer, Typechecker
+
+
+try:
+    from IPython.core.magic import line_magic, Magics, magics_class
+
+    @magics_class
+    class ChooseTypecheckerMagics(Magics):
+        @line_magic("jaxtyping.typechecker")
+        def typechecker(self, typechecker):
+            # remove old JaxtypingTransformer, if present
+            self.shell.ast_transformers = list(
+                filter(
+                    lambda x: not isinstance(x, JaxtypingTransformer),
+                    self.shell.ast_transformers,
+                )
+            )
+
+            # add new one
+            self.shell.ast_transformers.append(
+                JaxtypingTransformer(typechecker=Typechecker(typechecker))
+            )
+
+except Exception:
+    # Very broad exception-handling, as e.g. IPython will sometimes be
+    # present but fail to import for mysterious reasons.
+    pass
+
+
+def load_ipython_extension(ipython):
+    try:
+        ipython.register_magics(ChooseTypecheckerMagics)
+    except NameError:
+        raise NameError(
+            "ChooseTypecheckerMagics is not defined.\n\n"
+            + "You may be trying to use IPython extension without IPython installed."
+        )

jaxtyping/_pytree_type.py

@@ -19,11 +19,21 @@
 
 import functools as ft
 import typing
-from typing import Generic, TypeVar
+from typing import Any, Generic, TypeVar
 
 import jax.tree_util as jtu
 import typeguard
 
+from ._errors import AnnotationError
+from ._storage import (
+    clear_treeflatten_memo,
+    clear_treepath_memo,
+    get_shape_memo,
+    set_shape_memo,
+    set_treeflatten_memo,
+    set_treepath_memo,
+)
+
 
 _T = TypeVar("_T")
 
@@ -44,27 +54,133 @@ class _MetaPyTree(type):
     def __instancecheck__(cls, obj):
         if not hasattr(cls, "leaftype"):
             return True  # Just `isinstance(x, PyTree)`
+        # Handle beartype doing `isinstance(None, hint)` to check if
+        # is `instance`able.
+        if obj is None:
+            return True
 
-        # We could use `isinstance` here but that would fail for more complicated
-        # types, e.g. PyTree[Tuple[int]]. So at least internally we make a particular
-        # choice of typechecker.
-        #
-        # Deliberately not using @jaxtyped so that we share the same `memo` as whatever
-        # dynamic context we're currently in.
-        @typeguard.typechecked
-        def accepts_leaftype(x: cls.leaftype):
-            pass
+        single_memo, variadic_memo, pytree_memo, arg_memo = get_shape_memo()
+        single_memo_bak = single_memo.copy()
+        variadic_memo_bak = variadic_memo.copy()
+        pytree_memo_bak = pytree_memo.copy()
+        arg_memo_bak = arg_memo.copy()
+        try:
+            out = cls._check(obj, pytree_memo)
+        except Exception:
+            set_shape_memo(
+                single_memo_bak, variadic_memo_bak, pytree_memo_bak, arg_memo_bak
+            )
+            raise
+        if out:
+            return True
+        else:
+            set_shape_memo(
+                single_memo_bak, variadic_memo_bak, pytree_memo_bak, arg_memo_bak
+            )
+            return False
 
-        def is_leaftype(x):
-            try:
-                accepts_leaftype(x)
-            except _TypeCheckError:
+    def _check(cls, obj, pytree_memo):
+        if cls.leaftype is Any:
+
+            def is_flatten_leaftype(x):
                 return False
-            else:
+
+            def is_check_leaftype(x):
                 return True
 
-        leaves = jtu.tree_leaves(obj, is_leaf=is_leaftype)
-        return all(map(is_leaftype, leaves))
+        else:
+            # We could use `isinstance` here but that would fail for more complicated
+            # types, e.g. PyTree[tuple[int]]. So at least internally we make a
+            # particular choice of typechecker.
+            #
+            # Deliberately not using @jaxtyped so that we share the same `memo` as
+            # whatever dynamic context we're currently in.
+            @typeguard.typechecked
+            def accepts_leaftype(x: cls.leaftype):
+                pass
+
+            def is_leaftype(x):
+                try:
+                    accepts_leaftype(x)
+                except _TypeCheckError:
+                    return False
+                else:
+                    return True
+
+            is_flatten_leaftype = is_check_leaftype = is_leaftype
+
+        set_treeflatten_memo()
+        try:
+            leaves, structure = jtu.tree_flatten(obj, is_leaf=is_flatten_leaftype)
+        finally:
+            clear_treeflatten_memo()
+        if cls.structure is not None:
+            if cls.structure.isidentifier():
+                try:
+                    prev_structure = pytree_memo[cls.structure]
+                except KeyError:
+                    pytree_memo[cls.structure] = structure
+                else:
+                    if prev_structure != structure:
+                        return False
+            else:
+                named_pytree = 0
+                pieces = cls.structure.split()
+                if pieces[0] == "...":
+                    pieces = pieces[1:]
+                    prefix = False
+                    suffix = True
+                elif pieces[-1] == "...":
+                    pieces = pieces[:-1]
+                    prefix = True
+                    suffix = False
+                else:
+                    prefix = False
+                    suffix = False
+                for identifier in pieces:
+                    try:
+                        prev_structure = pytree_memo[identifier]
+                    except KeyError as e:
+                        raise AnnotationError(
+                            f"Cannot process composite structure '{cls.structure}' "
+                            f"as the structure name {identifier} has not been seen "
+                            "before."
+                        ) from e
+                    # Not using `PyTreeDef.compose` due to JAX bug #18218.
+                    prev_pytree = jtu.tree_unflatten(
+                        prev_structure, [0] * prev_structure.num_leaves
+                    )
+                    named_pytree = jtu.tree_map(lambda _: prev_pytree, named_pytree)
+                named_structure = jtu.tree_structure(named_pytree)
+                if prefix:
+                    dummy_pytree = jtu.tree_unflatten(structure, [0] * len(leaves))
+                    dummy_named = jtu.tree_unflatten(
+                        named_structure, [0] * named_structure.num_leaves
+                    )
+                    try:
+                        jtu.tree_map(lambda _, __: 0, dummy_named, dummy_pytree)
+                    except ValueError:
+                        return False
+                elif suffix:
+                    has_structure = lambda x: jtu.tree_structure(x) == named_structure
+                    dummy_pytree = jtu.tree_unflatten(structure, [0] * len(leaves))
+                    dummy_leaves = jtu.tree_leaves(dummy_pytree, is_leaf=has_structure)
+                    if any(not has_structure(x) for x in dummy_leaves):
+                        return False
+                else:
+                    if structure != named_structure:
+                        return False
+
+        try:
+            for leaf_index, leaf in enumerate(leaves):
+                if cls.structure is not None:
+                    set_treepath_memo(leaf_index, cls.structure)
+                if not is_check_leaftype(leaf):
+                    return False
+                clear_treepath_memo()
+        finally:
+            clear_treepath_memo()
+        return True
 
     # Can't return a generic (e.g. _FakePyTree[item]) because generic aliases don't do
     # the custom __instancecheck__ that we want.
@@ -75,10 +191,54 @@ class _MetaPyTree(type):
     # has __module__ "types", e.g. we get types.PyTree[int].
     @ft.lru_cache(maxsize=None)
     def __getitem__(cls, item):
-        name = str(_FakePyTree[item])
+        if isinstance(item, tuple):
+            if len(item) == 2:
 
-        class X(PyTree):
-            leaftype = item
+                class X(PyTree):
+                    leaftype = item[0]
+                    structure = item[1].strip()
+
+                if not isinstance(X.structure, str):
+                    raise ValueError(
+                        "The structure annotation `struct` in "
+                        "`jaxtyping.PyTree[leaftype, struct]` must be be a string, "
+                        f"e.g. `jaxtyping.PyTree[leaftype, 'T']`. Got '{X.structure}'."
+                    )
+                pieces = X.structure.split()
+                if len(pieces) == 0:
+                    raise ValueError(
+                        "The string `struct` in `jaxtyping.PyTree[leaftype, struct]` "
+                        "cannot be the empty string."
+                    )
+                for piece_index, piece in enumerate(pieces):
+                    if (piece_index == 0) or (piece_index == len(pieces) - 1):
+                        if piece == "...":
+                            continue
+                    if not piece.isidentifier():
+                        raise ValueError(
+                            "The string `struct` in "
+                            "`jaxtyping.PyTree[leaftype, struct]` must be be a "
+                            "whitespace-separated sequence of identifiers, e.g. "
+                            "`jaxtyping.PyTree[leaftype, 'T']` or "
+                            "`jaxtyping.PyTree[leaftype, 'foo bar']`.\n"
+                            "(Here, 'identifier' is used in the same sense as in "
+                            "regular Python, i.e. a valid variable name.)\n"
+                            f"Got piece '{piece}' in overall structure '{X.structure}'."
+                        )
+                name = str(_FakePyTree[item[0]])[:-1] + ', "' + item[1].strip() + '"]'
+            else:
+                raise ValueError(
+                    "The subscript `foo` in `jaxtyping.PyTree[foo]` must either be a "
+                    "leaf type, e.g. `PyTree[int]`, or a 2-tuple of leaf and "
+                    "structure, e.g. `PyTree[int, 'T']`. Received a tuple of length "
+                    f"{len(item)}."
+                )
+        else:
+            name = str(_FakePyTree[item])
+
+            class X(PyTree):
+                leaftype = item
+                structure = None
 
         X.__name__ = name
         X.__qualname__ = name
@@ -107,10 +267,56 @@ else:
     PyTree.__module__ = "jaxtyping"
 PyTree.__doc__ = """Represents a PyTree.
 
-Each PyTree is denoted by a type `PyTree[LeafType]`, such as `PyTree[int]` or
-`PyTree[Union[str, Float32[Array, "b c"]]]`.
+Annotations of the following sorts are supported:
+```python
+a: PyTree
+b: PyTree[LeafType]
+c: PyTree[LeafType, "T"]
+d: PyTree[LeafType, "S T"]
+e: PyTree[LeafType, "... T"]
+f: PyTree[LeafType, "T ..."]
+```
 
-You can leave off the `[...]`, in which case `PyTree` is simply a suggestively-named
-alternative to `Any`.
-([By definition all types are PyTrees.](https://jax.readthedocs.io/en/latest/pytrees.html))
+These correspond to:
+
+a. A plain `PyTree` can be used an annotation, in which case `PyTree` is simply a
+    suggestively-named alternative to `Any`.
+    ([By definition all types are PyTrees.](https://jax.readthedocs.io/en/latest/pytrees.html))
+
+b. `PyTree[LeafType]` denotes a PyTree all of whose leaves match `LeafType`. For
+    example, `PyTree[int]` or `PyTree[Union[str, Float32[Array, "b c"]]]`.
+
+c. A structure name can also be passed. In this case
+    `jax.tree_util.tree_structure(...)` will be called, and bound to the structure name.
+    This can be used to mark that multiple PyTrees all have the same structure:
+    ```python
+    def f(x: PyTree[int, "T"], y: PyTree[int, "T"]):
+        ...
+    ```
+    Structures are bound to names in the same way as array shape annotations, i.e.
+    within the thread-local dynamic context of a [`jaxtyping.jaxtyped`][] decorator.
+
+d. A composite structure can be declared. In this case the variable must have a PyTree
+    structure each to the composition of multiple previously-bound PyTree structures.
+    For example:
+    ```python
+    def f(x: PyTree[int, "T"], y: PyTree[int, "S"], z: PyTree[int, "S T"]):
+        ...
+
+    x = (1, 2)
+    y = {"key": 3}
+    z = {"key": (4, 5)}  # structure is the composition of the structures of `y` and `z`
+    f(x, y, z)
+    ```
+    When performing runtime type-checking, all the individual pieces must have already
+    been bound to structures, otherwise the composite structure check will throw an error.
+
+e. A structure can begin with a `...`, to denote that the lower levels of the PyTree
+    must match the declared structure, but the upper levels can be arbitrary. As in the
+    previous case, all named pieces must already have been seen and their structures
+    bound.
+
+f. A structure can end with a `...`, to denote that the PyTree must be a prefix of the
+    declared structure, but the lower levels can be arbitrary. As in the previous two
+    cases, all named pieces must already have been seen and their structures bound.
 """  # noqa: E501

jaxtyping/_storage.py

@@ -0,0 +1,176 @@
+# Copyright (c) 2022 Google LLC
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy of
+# this software and associated documentation files (the "Software"), to deal in
+# the Software without restriction, including without limitation the rights to
+# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
+# the Software, and to permit persons to whom the Software is furnished to do so,
+# subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
+# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
+# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+import threading
+from typing import Any, Optional
+
+from ._errors import AnnotationError
+
+
+_shape_storage = threading.local()
+
+
+def _has_shape_memo():
+    return hasattr(_shape_storage, "memo_stack") and len(_shape_storage.memo_stack) != 0
+
+
+def get_shape_memo():
+    if _has_shape_memo():
+        single_memo, variadic_memo, pytree_memo, arguments = _shape_storage.memo_stack[
+            -1
+        ]
+    else:
+        # `isinstance` happening outside any @jaxtyped decorators, e.g. at the
+        # global scope. In this case just create a temporary memo, since we're not
+        # going to be comparing against any stored values anyway.
+        single_memo = {}
+        variadic_memo = {}
+        pytree_memo = {}
+        arguments = {}
+    return single_memo, variadic_memo, pytree_memo, arguments
+
+
+def set_shape_memo(single_memo, variadic_memo, pytree_memo, arg_memo) -> None:
+    if _has_shape_memo():
+        _shape_storage.memo_stack[-1] = (
+            single_memo,
+            variadic_memo,
+            pytree_memo,
+            arg_memo,
+        )
+
+
+def push_shape_memo(arguments: dict[str, Any]):
+    try:
+        memo_stack = _shape_storage.memo_stack
+    except AttributeError:
+        # Can't be done when `_stack_storage` is created for reasons I forget.
+        memo_stack = _shape_storage.memo_stack = []
+    memos = ({}, {}, {}, arguments.copy())
+    memo_stack.append(memos)
+    return memos
+
+
+def pop_shape_memo() -> None:
+    _shape_storage.memo_stack.pop()
+
+
+def shape_str(memos) -> str:
+    """Gives debug information on the current state of jaxtyping's internal memos.
+    Used in type-checking error messages.
+
+    **Arguments:**
+
+    - `memos`: as returned by `get_shape_memo` or `push_shape_memo`.
+    """
+    single_memo, variadic_memo, pytree_memo, _ = memos
+    single_memo = {
+        name: size
+        for name, size in single_memo.items()
+        if not name.startswith("~~delete~~")
+    }
+    variadic_memo = {
+        name: shape
+        for name, (_, shape) in variadic_memo.items()
+        if not name.startswith("~~delete~~")
+    }
+    pieces = []
+    if len(single_memo) > 0 or len(variadic_memo) > 0:
+        pieces.append(
+            "The current values for each jaxtyping axis annotation are as follows."
+        )
+        for name, size in single_memo.items():
+            pieces.append(f"{name}={size}")
+        for name, shape in variadic_memo.items():
+            pieces.append(f"{name}={shape}")
+    if len(pytree_memo) > 0:
+        pieces.append(
+            "The current values for each jaxtyping PyTree structure annotation are as "
+            "follows."
+        )
+        for name, structure in pytree_memo.items():
+            pieces.append(f"{name}={structure}")
+    return "\n".join(pieces)
+
+
+def print_bindings():
+    """Prints the values of the current jaxtyping axis bindings. Intended for debugging.
+
+    That is, whilst doing runtime type checking, so that e.g. the `foo` and `bar` of
+    `Float[Array, "foo bar"]` are assigned values -- this function will print out those
+    values.
+
+    **Arguments:**
+
+    Nothing.
+
+    **Returns:**
+
+    Nothing.
+    """
+    print(shape_str(get_shape_memo()))
+
+
+_treepath_storage = threading.local()
+
+
+def clear_treepath_memo() -> None:
+    _treepath_storage.value = None
+
+
+def set_treepath_memo(index: Optional[int], structure: str) -> None:
+    if hasattr(_treepath_storage, "value") and _treepath_storage.value is not None:
+        raise AnnotationError(
+            "Cannot typecheck annotations of the form "
+            "`PyTree[PyTree[Shaped[Array, '?foo'], 'T'], 'S']` as it is ambiguous "
+            "which PyTree the `?` annotation refers to."
+        )
+    if index is None:
+        _treepath_storage.value = f"~~delete~~({structure}) "
+    else:
+        # Appears in error messages, so human-readable
+        _treepath_storage.value = f"(Leaf {index} in structure {structure}) "
+
+
+def get_treepath_memo() -> str:
+    if not hasattr(_treepath_storage, "value") or _treepath_storage.value is None:
+        raise AnnotationError(
+            "Cannot use `?` annotations, e.g. `Shaped[Array, '?foo']`, except "
+            "when contained with structured `PyTree` annotations, e.g. "
+            "`PyTree[Shaped[Array, '?foo'], 'T']`."
+        )
+    return _treepath_storage.value
+
+
+_treeflatten_storage = threading.local()
+
+
+def clear_treeflatten_memo() -> None:
+    _treeflatten_storage.value = False
+
+
+def set_treeflatten_memo():
+    _treeflatten_storage.value = True
+
+
+def get_treeflatten_memo():
+    try:
+        return _treeflatten_storage.value
+    except AttributeError:
+        return False

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "jaxtyping"
-version = "0.2.21"
+version = "0.2.26"
 description = "Type annotations and runtime checking for shape and dtype of JAX arrays, and PyTrees."
 readme = "README.md"
 requires-python ="~=3.9"
@@ -23,7 +23,7 @@ classifiers = [
     "Topic :: Scientific/Engineering :: Mathematics",
 ]
 urls = {repository = "https://github.com/google/jaxtyping" }
-dependencies = ["numpy>=1.20.0", "typeguard>=2.13.3", "typing_extensions>=3.7.4.1"]
+dependencies = ["numpy>=1.20.0", "typeguard==2.13.3"]
 entry-points = {pytest11 = {jaxtyping = "jaxtyping._pytest_plugin"}}
 
 [build-system]

test/conftest.py

@@ -41,6 +41,30 @@ def typecheck(request):
     return request.param
 
 
+@pytest.fixture(params=(False, True))
+def jaxtyp(request):
+    import jaxtyping
+
+    if request.param:
+        # New-style
+        # @jaxtyping.jaxtyped(typechecker=typechecker)
+        # def f(...)
+        return lambda typechecker: jaxtyping.jaxtyped(typechecker=typechecker)
+    else:
+        # Old-style
+        # @jaxtyping.jaxtyped
+        # @typechecker
+        # def f(...)
+        def impl(typechecker):
+            def decorator(fn):
+                with pytest.warns(match="As of jaxtyping version 0.2.24"):
+                    return jaxtyping.jaxtyped(typechecker(fn))
+
+            return decorator
+
+        return impl
+
+
 @pytest.fixture()
 def getkey():
     def _getkey():
@@ -48,3 +72,13 @@ def getkey():
         return jr.PRNGKey(random.randint(0, 2**31 - 1))
 
     return _getkey
+
+
+@pytest.fixture(scope="module")
+def beartype_or_skip():
+    yield pytest.importorskip("beartype")
+
+
+@pytest.fixture(scope="module")
+def typeguard_or_skip():
+    yield pytest.importorskip("typeguard")

test/import_hook_tester.py

@@ -0,0 +1,234 @@
+# Copyright (c) 2022 Google LLC
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy of
+# this software and associated documentation files (the "Software"), to deal in
+# the Software without restriction, including without limitation the rights to
+# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
+# the Software, and to permit persons to whom the Software is furnished to do so,
+# subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
+# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
+# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+import dataclasses
+
+import equinox as eqx
+import jax.numpy as jnp
+import pytest
+from helpers import ParamError, ReturnError
+
+import jaxtyping
+from jaxtyping import Float32, Int
+
+
+#
+# Test that functions get checked
+#
+
+
+def g(x: Float32[jnp.ndarray, " b"]):
+    pass
+
+
+g(jnp.array([1.0]))
+with pytest.raises(ParamError):
+    g(jnp.array(1))
+
+
+#
+# Test that Equinox modules get checked
+#
+
+
+# Dataclass `__init__`, no converter
+class Mod1(eqx.Module):
+    foo: int
+    bar: Float32[jnp.ndarray, " a"]
+
+
+Mod1(1, jnp.array([1.0]))
+with pytest.raises(ParamError):
+    Mod1(1.0, jnp.array([1.0]))
+with pytest.raises(ParamError):
+    Mod1(1, jnp.array(1.0))
+
+
+# Dataclass `__init__`, converter
+class Mod2(eqx.Module):
+    a: jnp.ndarray = eqx.field(converter=jnp.asarray)
+
+
+Mod2(1)  # This will fail unless we run typechecking after conversion
+
+
+class BadMod2(eqx.Module):
+    a: jnp.ndarray = eqx.field(converter=lambda x: x)
+
+
+with pytest.raises(ParamError):
+    BadMod2(1)
+with pytest.raises(ParamError):
+    BadMod2("asdf")
+
+
+# Custom `__init__`, no converter
+class Mod3(eqx.Module):
+    foo: int
+    bar: Float32[jnp.ndarray, " a"]
+
+    def __init__(self, foo: str, bar: Float32[jnp.ndarray, " a"]):
+        self.foo = int(foo)
+        self.bar = bar
+
+
+Mod3("1", jnp.array([1.0]))
+with pytest.raises(ParamError):
+    Mod3(1, jnp.array([1.0]))
+with pytest.raises(ParamError):
+    Mod3("1", jnp.array(1.0))
+
+
+# Custom `__init__`, converter
+class Mod4(eqx.Module):
+    a: Int[jnp.ndarray, ""] = eqx.field(converter=jnp.asarray)
+
+    def __init__(self, a: str):
+        self.a = int(a)
+
+
+Mod4("1")  # This will fail unless we run typechecking after conversion
+
+
+# Custom `__post_init__`, no converter
+class Mod5(eqx.Module):
+    foo: int
+    bar: Float32[jnp.ndarray, " a"]
+
+    def __post_init__(self):
+        pass
+
+
+Mod5(1, jnp.array([1.0]))
+with pytest.raises(ParamError):
+    Mod5(1.0, jnp.array([1.0]))
+with pytest.raises(ParamError):
+    Mod5(1, jnp.array(1.0))
+
+
+# Dataclass `__init__`, converter
+class Mod6(eqx.Module):
+    a: jnp.ndarray = eqx.field(converter=jnp.asarray)
+
+    def __post_init__(self):
+        pass
+
+
+Mod6(1)  # This will fail unless we run typechecking after conversion
+
+
+#
+# Test that dataclasses get checked
+#
+
+
+@dataclasses.dataclass
+class D:
+    foo: int
+    bar: Float32[jnp.ndarray, " a"]
+
+
+D(1, jnp.array([1.0]))
+with pytest.raises(ParamError):
+    D(1.0, jnp.array([1.0]))
+with pytest.raises(ParamError):
+    D(1, jnp.array(1.0))
+
+
+#
+# Test that methods get checked
+#
+
+
+class N(eqx.Module):
+    a: jnp.ndarray
+
+    def __init__(self, foo: str):
+        self.a = jnp.array(1)
+
+    def foo(self, x: jnp.ndarray):
+        pass
+
+    def bar(self) -> jnp.ndarray:
+        return self.a
+
+
+n = N("hi")
+with pytest.raises(ParamError):
+    N(123)
+with pytest.raises(ParamError):
+    n.foo("not_an_array_either")
+bad_n = eqx.tree_at(lambda x: x.a, n, "not_an_array")
+with pytest.raises(ReturnError):
+    bad_n.bar()
+
+
+#
+# Test that we don't get called in `super()`.
+#
+
+
+called = False
+
+
+class Base(eqx.Module):
+    x: int
+
+    def __init__(self):
+        self.x = "not an int"
+        global called
+        assert not called
+        called = True
+
+
+class Derived(Base):
+    def __init__(self):
+        assert not called
+        super().__init__()
+        assert called
+        self.x = 2
+
+
+Derived()
+
+
+#
+# Test that stringified type annotations work
+
+
+class Foo:
+    pass
+
+
+class Bar(eqx.Module):
+    x: type[Foo]
+    y: "type[Foo]"
+    # Note that this is the *only* kind of partially-stringified type annotation that
+    # is supported. This is for compatibility with older Equinox versions.
+    z: type["Foo"]
+
+
+Bar(Foo, Foo, Foo)
+
+with pytest.raises(ParamError):
+    Bar(1, Foo, Foo)
+
+# Record that we've finished our checks successfully
+
+jaxtyping._test_import_hook_counter += 1

test/import_hook_tester_beartype.py

@@ -1,62 +0,0 @@
-# Copyright (c) 2022 Google LLC
-#
-# Permission is hereby granted, free of charge, to any person obtaining a copy of
-# this software and associated documentation files (the "Software"), to deal in
-# the Software without restriction, including without limitation the rights to
-# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
-# the Software, and to permit persons to whom the Software is furnished to do so,
-# subject to the following conditions:
-#
-# The above copyright notice and this permission notice shall be included in all
-# copies or substantial portions of the Software.
-#
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
-# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
-# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
-# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-import dataclasses
-
-import equinox as eqx
-import jax.numpy as jnp
-import pytest
-
-from jaxtyping import Float32
-
-from .helpers import ParamError
-
-
-def g(x: Float32[jnp.ndarray, " b"]):
-    pass
-
-
-g(jnp.array([1.0]))
-with pytest.raises(ParamError):
-    g(jnp.array(1))
-
-
-class M(eqx.Module):
-    foo: int
-    bar: Float32[jnp.ndarray, " a"]
-
-
-M(1, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    M(1.0, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    M(1, jnp.array(1.0))
-
-
-@dataclasses.dataclass
-class D:
-    foo: int
-    bar: Float32[jnp.ndarray, " a"]
-
-
-D(1, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    D(1.0, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    D(1, jnp.array(1.0))

test/import_hook_tester_beartype_full.py

@@ -1,62 +0,0 @@
-# Copyright (c) 2022 Google LLC
-#
-# Permission is hereby granted, free of charge, to any person obtaining a copy of
-# this software and associated documentation files (the "Software"), to deal in
-# the Software without restriction, including without limitation the rights to
-# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
-# the Software, and to permit persons to whom the Software is furnished to do so,
-# subject to the following conditions:
-#
-# The above copyright notice and this permission notice shall be included in all
-# copies or substantial portions of the Software.
-#
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
-# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
-# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
-# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-import dataclasses
-
-import equinox as eqx
-import jax.numpy as jnp
-import pytest
-
-from jaxtyping import Float32
-
-from .helpers import ParamError
-
-
-def g(x: Float32[jnp.ndarray, " b"]):
-    pass
-
-
-g(jnp.array([1.0]))
-with pytest.raises(ParamError):
-    g(jnp.array(1))
-
-
-class M(eqx.Module):
-    foo: int
-    bar: Float32[jnp.ndarray, " a"]
-
-
-M(1, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    M(1.0, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    M(1, jnp.array(1.0))
-
-
-@dataclasses.dataclass
-class D:
-    foo: int
-    bar: Float32[jnp.ndarray, " a"]
-
-
-D(1, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    D(1.0, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    D(1, jnp.array(1.0))

test/import_hook_tester_beartype_old.py

@@ -1,62 +0,0 @@
-# Copyright (c) 2022 Google LLC
-#
-# Permission is hereby granted, free of charge, to any person obtaining a copy of
-# this software and associated documentation files (the "Software"), to deal in
-# the Software without restriction, including without limitation the rights to
-# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
-# the Software, and to permit persons to whom the Software is furnished to do so,
-# subject to the following conditions:
-#
-# The above copyright notice and this permission notice shall be included in all
-# copies or substantial portions of the Software.
-#
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
-# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
-# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
-# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-import dataclasses
-
-import equinox as eqx
-import jax.numpy as jnp
-import pytest
-
-from jaxtyping import Float32
-
-from .helpers import ParamError
-
-
-def g(x: Float32[jnp.ndarray, " b"]):
-    pass
-
-
-g(jnp.array([1.0]))
-with pytest.raises(ParamError):
-    g(jnp.array(1))
-
-
-class M(eqx.Module):
-    foo: int
-    bar: Float32[jnp.ndarray, " a"]
-
-
-M(1, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    M(1.0, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    M(1, jnp.array(1.0))
-
-
-@dataclasses.dataclass
-class D:
-    foo: int
-    bar: Float32[jnp.ndarray, " a"]
-
-
-D(1, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    D(1.0, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    D(1, jnp.array(1.0))

test/import_hook_tester_broken_checker.py

@@ -1,62 +0,0 @@
-# Copyright (c) 2022 Google LLC
-#
-# Permission is hereby granted, free of charge, to any person obtaining a copy of
-# this software and associated documentation files (the "Software"), to deal in
-# the Software without restriction, including without limitation the rights to
-# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
-# the Software, and to permit persons to whom the Software is furnished to do so,
-# subject to the following conditions:
-#
-# The above copyright notice and this permission notice shall be included in all
-# copies or substantial portions of the Software.
-#
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
-# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
-# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
-# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-import dataclasses
-
-import equinox as eqx
-import jax.numpy as jnp
-import pytest
-
-from jaxtyping import Float32
-
-from .helpers import ParamError
-
-
-def g(x: Float32[jnp.ndarray, " b"]):
-    pass
-
-
-g(jnp.array([1.0]))
-with pytest.raises(ParamError):
-    g(jnp.array(1))
-
-
-class M(eqx.Module):
-    foo: int
-    bar: Float32[jnp.ndarray, " a"]
-
-
-M(1, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    M(1.0, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    M(1, jnp.array(1.0))
-
-
-@dataclasses.dataclass
-class D:
-    foo: int
-    bar: Float32[jnp.ndarray, " a"]
-
-
-D(1, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    D(1.0, jnp.array([1.0]))
-with pytest.raises(ParamError):
-    D(1, jnp.array(1.0))

test/import_hook_tester_transitive/__init__.py

@@ -1,20 +0,0 @@
-# Copyright (c) 2022 Google LLC
-#
-# Permission is hereby granted, free of charge, to any person obtaining a copy of
-# this software and associated documentation files (the "Software"), to deal in
-# the Software without restriction, including without limitation the rights to
-# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
-# the Software, and to permit persons to whom the Software is furnished to do so,
-# subject to the following conditions:
-#
-# The above copyright notice and this permission notice shall be included in all
-# copies or substantial portions of the Software.
-#
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
-# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
-# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
-# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-from . import another_file  # noqa: F401

test/import_hook_tester_transitive/another_file.py

@@ -1,48 +0,0 @@
-# Copyright (c) 2022 Google LLC
-#
-# Permission is hereby granted, free of charge, to any person obtaining a copy of
-# this software and associated documentation files (the "Software"), to deal in
-# the Software without restriction, including without limitation the rights to
-# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
-# the Software, and to permit persons to whom the Software is furnished to do so,
-# subject to the following conditions:
-#
-# The above copyright notice and this permission notice shall be included in all
-# copies or substantial portions of the Software.
-#
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
-# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
-# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
-# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-import jax.numpy as jnp
-import pytest
-
-from jaxtyping import Float32
-
-from ..helpers import ParamError
-
-
-def g(x: Float32[jnp.ndarray, " b"]):
-    pass
-
-
-g(jnp.array([1.0]))
-with pytest.raises(ParamError):
-    g(jnp.array(1))
-
-
-# Typeguard 3.0 no longer supports this
-#
-# class M(eqx.Module):
-#     foo: int
-#     bar: Float32[jnp.ndarray, " a"]
-
-
-# M(1, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     M(1.0, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     M(1, jnp.array(1.0))

test/import_hook_tester_typeguard.py

@@ -1,63 +0,0 @@
-# Copyright (c) 2022 Google LLC
-#
-# Permission is hereby granted, free of charge, to any person obtaining a copy of
-# this software and associated documentation files (the "Software"), to deal in
-# the Software without restriction, including without limitation the rights to
-# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
-# the Software, and to permit persons to whom the Software is furnished to do so,
-# subject to the following conditions:
-#
-# The above copyright notice and this permission notice shall be included in all
-# copies or substantial portions of the Software.
-#
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
-# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
-# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
-# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-
-import jax.numpy as jnp
-import pytest
-
-from jaxtyping import Float32
-
-from .helpers import ParamError
-
-
-def g(x: Float32[jnp.ndarray, " b"]):
-    pass
-
-
-g(jnp.array([1.0]))
-with pytest.raises(ParamError):
-    g(jnp.array(1))
-
-
-# Typeguard 3.0 no longer supports this.
-#
-# class M(eqx.Module):
-#     foo: int
-#     bar: Float32[jnp.ndarray, " a"]
-#
-#
-# M(1, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     M(1.0, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     M(1, jnp.array(1.0))
-#
-#
-#
-# @dataclasses.dataclass
-# class D:
-#     foo: int
-#     bar: Float32[jnp.ndarray, " a"]
-#
-#
-# D(1, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     D(1.0, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     D(1, jnp.array(1.0))

test/import_hook_tester_typeguard_old.py

@@ -1,63 +0,0 @@
-# Copyright (c) 2022 Google LLC
-#
-# Permission is hereby granted, free of charge, to any person obtaining a copy of
-# this software and associated documentation files (the "Software"), to deal in
-# the Software without restriction, including without limitation the rights to
-# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
-# the Software, and to permit persons to whom the Software is furnished to do so,
-# subject to the following conditions:
-#
-# The above copyright notice and this permission notice shall be included in all
-# copies or substantial portions of the Software.
-#
-# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
-# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
-# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
-# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-
-import jax.numpy as jnp
-import pytest
-
-from jaxtyping import Float32
-
-from .helpers import ParamError
-
-
-def g(x: Float32[jnp.ndarray, " b"]):
-    pass
-
-
-g(jnp.array([1.0]))
-with pytest.raises(ParamError):
-    g(jnp.array(1))
-
-
-# Typeguard 3.0 no longer supports this.
-#
-# class M(eqx.Module):
-#     foo: int
-#     bar: Float32[jnp.ndarray, " a"]
-#
-#
-# M(1, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     M(1.0, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     M(1, jnp.array(1.0))
-#
-#
-#
-# @dataclasses.dataclass
-# class D:
-#     foo: int
-#     bar: Float32[jnp.ndarray, " a"]
-#
-#
-# D(1, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     D(1.0, jnp.array([1.0]))
-# with pytest.raises(ParamError):
-#     D(1, jnp.array(1.0))

test/requirements.txt

@@ -1,6 +1,9 @@
 beartype
 cloudpickle
 equinox
+IPython
 jaxlib
 pytest
+pytest-asyncio
+tensorflow
 typeguard<3

test/test_all_importable.py

@@ -0,0 +1,47 @@
+# We have some pretty complicated semantics in `__init__.py`.
+# Here we check that we didn't miss one of them on our runtime branch.
+def test_all_importable():
+    # Ordered according to their appearance in the documentation.
+    from jaxtyping import (  # noqa: I001
+        Shaped,  # noqa: F401
+        Bool,  # noqa: F401
+        Key,  # noqa: F401
+        Num,  # noqa: F401
+        Inexact,  # noqa: F401
+        Float,  # noqa: F401
+        BFloat16,  # noqa: F401
+        Float16,  # noqa: F401
+        Float32,  # noqa: F401
+        Float64,  # noqa: F401
+        Complex,  # noqa: F401
+        Complex64,  # noqa: F401
+        Complex128,  # noqa: F401
+        Integer,  # noqa: F401
+        UInt,  # noqa: F401
+        UInt4,  # noqa: F401
+        UInt8,  # noqa: F401
+        UInt16,  # noqa: F401
+        UInt32,  # noqa: F401
+        UInt64,  # noqa: F401
+        Int,  # noqa: F401
+        Int4,  # noqa: F401
+        Int8,  # noqa: F401
+        Int16,  # noqa: F401
+        Int32,  # noqa: F401
+        Int64,  # noqa: F401
+        Real,  # noqa: F401
+        Array,  # noqa: F401
+        ArrayLike,  # noqa: F401
+        Scalar,  # noqa: F401
+        ScalarLike,  # noqa: F401
+        PRNGKeyArray,  # noqa: F401
+        PyTreeDef,  # noqa: F401
+        PyTree,  # noqa: F401
+        jaxtyped,  # noqa: F401
+        install_import_hook,  # noqa: F401
+        AbstractArray,  # noqa: F401
+        AbstractDtype,  # noqa: F401
+        print_bindings,  # noqa: F401
+        get_array_name_format,  # noqa: F401
+        set_array_name_format,  # noqa: F401
+    )

test/test_array.py

@@ -17,6 +17,7 @@
 # IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 # CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
+import dataclasses as dc
 import sys
 from typing import get_args, get_origin, Union
 
@@ -24,25 +25,33 @@ import jax.numpy as jnp
 import jax.random as jr
 import numpy as np
 import pytest
-import torch
+
+
+try:
+    import torch
+except ImportError:
+    torch = None
 
 from jaxtyping import (
     AbstractDtype,
+    AnnotationError,
     Array,
     ArrayLike,
+    Bool,
     Float,
     Float32,
     jaxtyped,
+    Key,
     PRNGKeyArray,
+    Scalar,
     Shaped,
 )
 
 from .helpers import ParamError, ReturnError
 
 
-def test_basic(typecheck):
-    @jaxtyped
-    @typecheck
+def test_basic(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
     def g(x: Shaped[Array, "..."]):
         pass
 
@@ -63,6 +72,7 @@ def test_dtypes():
         Float64,
         Inexact,
         Int,
+        Int4,
         Int8,
         Int16,
         Int32,
@@ -70,6 +80,7 @@ def test_dtypes():
         Num,
         Shaped,
         UInt,
+        UInt4,
         UInt8,
         UInt16,
         UInt32,
@@ -81,16 +92,14 @@ def test_dtypes():
             assert key == val.__name__
 
 
-def test_return(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_return(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float[Array, "b c"]) -> Float[Array, "c b"]:
         return jnp.transpose(x)
 
     g(jr.normal(getkey(), (3, 4)))
 
-    @jaxtyped
-    @typecheck
+    @jaxtyp(typecheck)
     def h(x: Float[Array, "b c"]) -> Float[Array, "b c"]:
         return jnp.transpose(x)
 
@@ -98,9 +107,8 @@ def test_return(typecheck, getkey):
         h(jr.normal(getkey(), (3, 4)))
 
 
-def test_two_args(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_two_args(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Shaped[Array, "b c"], y: Shaped[Array, "c d"]):
         return x @ y
 
@@ -108,8 +116,7 @@ def test_two_args(typecheck, getkey):
     with pytest.raises(ParamError):
         g(jr.normal(getkey(), (3, 4)), jr.normal(getkey(), (5, 4)))
 
-    @jaxtyped
-    @typecheck
+    @jaxtyp(typecheck)
     def h(x: Shaped[Array, "b c"], y: Shaped[Array, "c d"]) -> Shaped[Array, "b d"]:
         return x @ y
 
@@ -118,15 +125,16 @@ def test_two_args(typecheck, getkey):
         h(jr.normal(getkey(), (3, 4)), jr.normal(getkey(), (5, 4)))
 
 
-def test_any_dtype(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_any_dtype(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Shaped[Array, "a b"]) -> Shaped[Array, "a b"]:
         return x
 
     g(jr.normal(getkey(), (3, 4)))
     g(jnp.array([[True, False]]))
+    g(jnp.array([[1, 2], [3, 4]], dtype=jnp.int4))
     g(jnp.array([[1, 2], [3, 4]], dtype=jnp.int8))
+    g(jnp.array([[1, 2], [3, 4]], dtype=jnp.uint4))
     g(jnp.array([[1, 2], [3, 4]], dtype=jnp.uint16))
     g(jr.normal(getkey(), (3, 4), dtype=jnp.complex128))
     g(jr.normal(getkey(), (3, 4), dtype=jnp.bfloat16))
@@ -135,14 +143,12 @@ def test_any_dtype(typecheck, getkey):
         g(jr.normal(getkey(), (1,)))
 
 
-def test_nested_jaxtyped(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_nested_jaxtyped(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float32[Array, "b c"], transpose: bool) -> Float32[Array, "c b"]:
         return h(x, transpose)
 
-    @jaxtyped
-    @typecheck
+    @jaxtyp(typecheck)
     def h(x: Float32[Array, "c b"], transpose: bool) -> Float32[Array, "b c"]:
         if transpose:
             return jnp.transpose(x)
@@ -156,9 +162,8 @@ def test_nested_jaxtyped(typecheck, getkey):
         g(jr.normal(getkey(), (2, 3)), False)
 
 
-def test_nested_nojaxtyped(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_nested_nojaxtyped(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float32[Array, "b c"]):
         return h(x)
 
@@ -170,9 +175,8 @@ def test_nested_nojaxtyped(typecheck, getkey):
         g(jr.normal(getkey(), (2, 3)))
 
 
-def test_isinstance(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_isinstance(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float32[Array, "b c"]) -> Float32[Array, " z"]:
         y = jnp.transpose(x)
         assert isinstance(y, Float32[Array, "c b"])
@@ -186,9 +190,8 @@ def test_isinstance(typecheck, getkey):
     g(jr.normal(getkey(), (2, 3)))
 
 
-def test_fixed(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_fixed(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(
         x: Float32[Array, "4 5 foo"], y: Float32[Array, " foo"]
     ) -> Float32[Array, "4 5"]:
@@ -203,9 +206,8 @@ def test_fixed(typecheck, getkey):
         g(c, b)
 
 
-def test_anonymous(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_anonymous(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float32[Array, "foo _"], y: Float32[Array, " _"]):
         pass
 
@@ -214,9 +216,8 @@ def test_anonymous(typecheck, getkey):
     g(a, b)
 
 
-def test_named_variadic(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_named_variadic(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(
         x: Float32[Array, "*batch foo"],
         y: Float32[Array, " *batch"],
@@ -239,8 +240,7 @@ def test_named_variadic(typecheck, getkey):
     with pytest.raises(ParamError):
         g(a2, b1, c)
 
-    @jaxtyped
-    @typecheck
+    @jaxtyp(typecheck)
     def h(x: Float32[Array, " foo *batch"], y: Float32[Array, " foo *batch bar"]):
         pass
 
@@ -254,9 +254,8 @@ def test_named_variadic(typecheck, getkey):
         h(b, c)
 
 
-def test_anonymous_variadic(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_anonymous_variadic(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float32[Array, "... foo"], y: Float32[Array, " foo"]):
         pass
 
@@ -276,9 +275,8 @@ def test_anonymous_variadic(typecheck, getkey):
         g(a3, c)
 
 
-def test_broadcast_fixed(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_broadcast_fixed(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float32[Array, "#4"]):
         pass
 
@@ -289,9 +287,8 @@ def test_broadcast_fixed(typecheck, getkey):
         g(jr.normal(getkey(), (3,)))
 
 
-def test_broadcast_named(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_broadcast_named(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float32[Array, " #foo"], y: Float32[Array, " #foo"]):
         pass
 
@@ -313,9 +310,8 @@ def test_broadcast_named(typecheck, getkey):
         g(b, a)
 
 
-def test_broadcast_variadic_named(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_broadcast_variadic_named(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: Float32[Array, " *#foo"], y: Float32[Array, " *#foo"]):
         pass
 
@@ -372,29 +368,71 @@ def test_broadcast_variadic_named(typecheck, getkey):
         g(o, a)
 
 
+def test_variadic_mixed_broadcast(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
+    def f(x: Float[Array, " *foo"], y: Float[Array, " #*foo"]):
+        pass
+
+    a = jr.normal(getkey(), (3, 4))
+    b = jr.normal(getkey(), (5,))
+    with pytest.raises(ParamError):
+        f(a, b)
+
+    c = jr.normal(getkey(), (7, 3, 2))
+    d = jr.normal(getkey(), (1, 2))
+    f(c, d)
+
+
+def test_variadic_mixed_broadcast2(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
+    def f(x: Float[Array, " *#foo"], y: Float[Array, " *foo"]):
+        pass
+
+    a = jr.normal(getkey(), (3, 4))
+    b = jr.normal(getkey(), (5,))
+    with pytest.raises(ParamError):
+        f(a, b)
+
+    c = jr.normal(getkey(), (1, 2))
+    d = jr.normal(getkey(), (7, 3, 2))
+    f(c, d)
+
+
+def test_variadic_mixed_broadcast3(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
+    def f(
+        x: Float[Array, "*B L D"],
+        *,
+        y: Float[Array, "*#B J d"],
+        z: Bool[Array, "*B L J"],
+    ) -> Float[Array, "*B L D"]:
+        return x
+
+    x = jr.normal(getkey(), (2, 7, 3, 2, 2))
+    y = jr.bernoulli(getkey(), shape=(2, 7, 3, 2, 2))
+    z = jr.normal(getkey(), (2, 7, 1, 2, 2))
+    f(x, y=z, z=y)
+
+
 def test_no_commas():
     with pytest.raises(ValueError):
         Float32[Array, "foo, bar"]
 
 
-def test_symbolic(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_symbolic(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def make_slice(x: Float32[Array, " dim"]) -> Float32[Array, " dim-1"]:
         return x[1:]
 
-    @jaxtyped
-    @typecheck
+    @jaxtyp(typecheck)
     def cat(x: Float32[Array, " dim"]) -> Float32[Array, " 2*dim"]:
         return jnp.concatenate([x, x])
 
-    @jaxtyped
-    @typecheck
+    @jaxtyp(typecheck)
     def bad_make_slice(x: Float32[Array, " dim"]) -> Float32[Array, " dim-1"]:
         return x
 
-    @jaxtyped
-    @typecheck
+    @jaxtyp(typecheck)
     def bad_cat(x: Float32[Array, " dim"]) -> Float32[Array, " 2*dim"]:
         return jnp.concatenate([x, x, x])
 
@@ -414,17 +452,64 @@ def test_symbolic(typecheck, getkey):
         bad_cat(x)
 
 
-def test_incomplete_symbolic(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_incomplete_symbolic(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def foo(x: Float32[Array, " 2*dim"]):
         pass
 
     x = jr.normal(getkey(), (4,))
-    with pytest.raises(NameError):
+    with pytest.raises(AnnotationError):
         foo(x)
 
 
+def test_deferred_symbolic_good(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def foo(dim: int, fill: Float[Array, ""]) -> Float[Array, " {dim}"]:
+        return jnp.full((dim,), fill)
+
+    class A:
+        size = 5
+
+        @jaxtyp(typecheck)
+        def bar(self, fill: Float[Array, ""]) -> Float[Array, " {self.size}"]:
+            return jnp.full((self.size,), fill)
+
+    foo(3, jnp.array(0.0))
+    A().bar(jnp.array(0.0))
+
+
+def test_deferred_symbolic_bad(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def foo(dim: int, fill: Float[Array, ""]) -> Float[Array, " {dim-1}"]:
+        return jnp.full((dim,), fill)
+
+    class A:
+        size = 5
+
+        @jaxtyp(typecheck)
+        def bar(self, fill: Float[Array, ""]) -> Float[Array, " {self.size}-1"]:
+            return jnp.full((self.size,), fill)
+
+    with pytest.raises(ReturnError):
+        foo(3, jnp.array(0.0))
+
+    with pytest.raises(ReturnError):
+        A().bar(jnp.array(0.0))
+
+
+def test_deferred_symbolic_dataclass(typecheck):
+    @jaxtyped(typechecker=typecheck)
+    @dc.dataclass
+    class A:
+        value: int
+        array: Float[Array, " {self.value}"]
+
+    A(3, jnp.zeros(3))
+
+    with pytest.raises(ParamError):
+        A(3, jnp.zeros(4))
+
+
 def test_arraylike(typecheck, getkey):
     floatlike1 = Float32[ArrayLike, ""]
     floatlike2 = Float[ArrayLike, ""]
@@ -436,22 +521,18 @@ def test_arraylike(typecheck, getkey):
     assert set(get_args(floatlike1)) == {
         Float32[Array, ""],
         Float32[np.ndarray, ""],
-        Float32[np.bool_, ""],
         Float32[np.number, ""],
         float,
     }
     assert set(get_args(floatlike2)) == {
         Float[Array, ""],
         Float[np.ndarray, ""],
-        Float[np.bool_, ""],
         Float[np.number, ""],
         float,
     }
     assert set(get_args(floatlike3)) == {
         Float32[Array, "4"],
         Float32[np.ndarray, "4"],
-        Float32[np.bool_, "4"],
-        Float32[np.number, "4"],
     }
 
     shaped1 = Shaped[ArrayLike, ""]
@@ -471,15 +552,15 @@ def test_arraylike(typecheck, getkey):
     assert set(get_args(shaped2)) == {
         Shaped[Array, "4"],
         Shaped[np.ndarray, "4"],
-        Shaped[np.bool_, "4"],
-        Shaped[np.number, "4"],
     }
 
 
 def test_subclass():
     assert issubclass(Float[Array, ""], Array)
     assert issubclass(Float[np.ndarray, ""], np.ndarray)
-    assert issubclass(Float[torch.Tensor, ""], torch.Tensor)
+
+    if torch is not None:
+        assert issubclass(Float[torch.Tensor, ""], torch.Tensor)
 
 
 def test_ignored_names():
@@ -523,14 +604,13 @@ def test_py310_unions():
     assert isinstance(x, get_args(y))
 
 
-def test_key(typecheck):
-    @jaxtyped
-    @typecheck
+def test_key(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
     def f(x: PRNGKeyArray):
         pass
 
-    x = jr.PRNGKey(0)
-    f(x)
+    f(jr.key(0))
+    f(jr.PRNGKey(0))
 
     with pytest.raises(ParamError):
         f(object())
@@ -542,7 +622,31 @@ def test_key(typecheck):
         f(jnp.array(3.0))
 
 
-def test_extension(typecheck, getkey):
+def test_key_dtype(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f1(x: Key[Array, ""]):
+        pass
+
+    @jaxtyp(typecheck)
+    def f2(x: Key[Scalar, ""]):
+        pass
+
+    for f in (f1, f2):
+        f(jr.key(0))
+
+        with pytest.raises(ParamError):
+            f(jr.PRNGKey(0))
+        with pytest.raises(ParamError):
+            f(object())
+        with pytest.raises(ParamError):
+            f(1)
+        with pytest.raises(ParamError):
+            f(jnp.array(3))
+        with pytest.raises(ParamError):
+            f(jnp.array(3.0))
+
+
+def test_extension(jaxtyp, typecheck, getkey):
     X = Shaped[Array, "a b"]
     Y = Shaped[X, "c d"]
     Z = Shaped[Array, "c d a b"]
@@ -551,8 +655,7 @@ def test_extension(typecheck, getkey):
     X = Float[Array, "a"]
     Y = Float[X, "b"]
 
-    @jaxtyped
-    @typecheck
+    @jaxtyp(typecheck)
     def f(a: X, b: Y):
         ...
 
@@ -575,3 +678,17 @@ def test_extension(typecheck, getkey):
     g(jr.split(jr.PRNGKey(0)))
     with pytest.raises(ParamError):
         g(jr.split(jr.PRNGKey(0), 3))
+
+
+def test_scalar_variadic_dim():
+    assert Float[float, "..."] is float
+    assert Float[float, "#*shape"] is float
+
+    # This one is a bit weird -- it should really also assert that shape==(), but we
+    # don't implement that.
+    assert Float[float, "*shape"] is float
+
+
+def test_scalar_dtype_mismatch():
+    with pytest.raises(ValueError):
+        Float[bool, "..."]

test/test_decorator.py

@@ -1,52 +1,58 @@
 import abc
 
-from jaxtyping import jaxtyped
+import jax.numpy as jnp
+import jax.random as jr
+import pytest
+
+from jaxtyping import Array, Float, jaxtyped, print_bindings
+
+from .helpers import ParamError, ReturnError
 
 
 class M(metaclass=abc.ABCMeta):
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     def f(self):
         ...
 
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     @classmethod
     def g1(cls):
         return 3
 
     @classmethod
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     def g2(cls):
         return 4
 
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     @staticmethod
     def h1():
         return 3
 
     @staticmethod
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     def h2():
         return 4
 
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     @abc.abstractmethod
     def i1(self):
         ...
 
     @abc.abstractmethod
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     def i2(self):
         ...
 
 
 class N:
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     @property
     def j1(self):
         return 3
 
     @property
-    @jaxtyped
+    @jaxtyped(typechecker=None)
     def j2(self):
         return 4
 
@@ -75,3 +81,103 @@ def test_abstractmethod():
 def test_property():
     assert N().j1 == 3
     assert N().j2 == 4
+
+
+def test_context(getkey):
+    a = jr.normal(getkey(), (3, 4))
+    b = jr.normal(getkey(), (5,))
+    with jaxtyped("context"):
+        assert isinstance(a, Float[Array, "foo bar"])
+        assert not isinstance(b, Float[Array, "foo"])
+    assert isinstance(a, Float[Array, "foo bar"])
+    assert isinstance(b, Float[Array, "foo"])
+
+
+def test_varargs(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f(*args):
+        pass
+
+    f(1, 2)
+
+
+def test_varkwargs(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f(**kwargs):
+        pass
+
+    f(a=1, b=2)
+
+
+def test_defaults(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f(x: int, y=1):
+        pass
+
+    f(1)
+
+
+class _GlobalFoo:
+    pass
+
+
+def test_global_stringified_annotation(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f(x: "_GlobalFoo") -> "_GlobalFoo":
+        return x
+
+    f(_GlobalFoo())
+
+    @jaxtyp(typecheck)
+    def g(x: int) -> "_GlobalFoo":
+        return x
+
+    @jaxtyp(typecheck)
+    def h(x: "_GlobalFoo") -> int:
+        return x
+
+    with pytest.raises(ReturnError):
+        g(1)
+
+    with pytest.raises(ParamError):
+        h(1)
+
+
+# This test does not use `jaxtyp(typecheck)` because typeguard does some evil stack
+# frame introspection to try and grab local variables.
+def test_local_stringified_annotation(typecheck):
+    class LocalFoo:
+        pass
+
+    @jaxtyped(typechecker=typecheck)
+    def f(x: "LocalFoo") -> "LocalFoo":
+        return x
+
+    f(LocalFoo())
+
+    with pytest.warns(match="As of jaxtyping version 0.2.24"):
+
+        @jaxtyped
+        @typecheck
+        def g(x: "LocalFoo") -> "LocalFoo":
+            return x
+
+    g(LocalFoo())
+
+    # We don't check that errors are raised if it goes wrong, since we can't usually
+    # resolve local type annotations at runtime. Best we can hope for is not to raise
+    # a spurious error about not being able to find the type.
+
+
+def test_print_bindings(typecheck, capfd):
+    @jaxtyped(typechecker=typecheck)
+    def f(x: Float[Array, "foo bar"]):
+        print_bindings()
+
+    capfd.readouterr()
+    f(jnp.zeros((3, 4)))
+    text, _ = capfd.readouterr()
+    assert text == (
+        "The current values for each jaxtyping axis annotation are as follows."
+        "\nfoo=3\nbar=4\n"
+    )

test/test_equals.py

@@ -0,0 +1,35 @@
+from typing import Tuple, Union
+
+import pytest
+
+from jaxtyping import (
+    Array,
+    Float,
+    Float32,
+    Integer,
+    PRNGKeyArray,
+    PyTree,
+    Shaped,
+)
+
+
+@pytest.mark.parametrize(
+    "make_fn",
+    [
+        lambda: Float[Array, "4"],
+        lambda: Float32[Array, ""],
+        lambda: Integer[Array, "1 2 3"],
+        lambda: Shaped[PRNGKeyArray, "2"],
+        lambda: Float[float, "#*shape"],
+        lambda: PyTree[int],
+        lambda: PyTree[Float[Array, ""]],
+        lambda: PyTree[Float32[Array, "*m b c"]],
+        lambda: PyTree[PyTree[Float32[Array, "1 2 b *"]]],
+        lambda: PyTree[Union[str, Float32[Array, "1"]]],
+        lambda: PyTree[
+            Tuple[int, float, Float[Array, ""], PyTree[Union[Float[Array, ""], float]]]
+        ],
+    ],
+)
+def test_equals(make_fn):
+    assert make_fn() == make_fn()

test/test_generators.py

@@ -0,0 +1,88 @@
+from typing import AsyncIterator, Iterator
+
+import jax.numpy as jnp
+import pytest
+
+from jaxtyping import Array, Float, Shaped
+
+from .helpers import ParamError
+
+
+try:
+    import torch
+except ImportError:
+    torch = None
+
+
+def test_generators_simple(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def gen(x: Float[Array, "*"]) -> Iterator[Float[Array, "*"]]:
+        yield x
+
+    @jaxtyp(typecheck)
+    def foo():
+        next(gen(jnp.zeros(2)))
+        next(gen(jnp.zeros((3, 4))))
+
+    foo()
+
+
+def test_generators_return_no_annotations(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def gen(x: Float[Array, "*"]):
+        yield x
+
+    @jaxtyp(typecheck)
+    def foo():
+        next(gen(jnp.zeros(2)))
+        next(gen(jnp.zeros((3, 4))))
+
+    foo()
+
+
+@pytest.mark.asyncio
+async def test_async_generators_simple(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    async def gen(x: Float[Array, "*"]) -> AsyncIterator[Float[Array, "*"]]:
+        yield x
+
+    @jaxtyp(typecheck)
+    async def foo():
+        async for _ in gen(jnp.zeros(2)):
+            pass
+        async for _ in gen(jnp.zeros((3, 4))):
+            pass
+
+    await foo()
+
+
+def test_generators_dont_modify_same_annotations(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def g(x: Float[Array, "1"]) -> Iterator[Float[Array, "1"]]:
+        yield x
+
+    @jaxtyp(typecheck)
+    def m(x: Float[Array, "1"]) -> Float[Array, "1"]:
+        return x
+
+    with pytest.raises(ParamError):
+        next(g(jnp.zeros(2)))
+    with pytest.raises(ParamError):
+        m(jnp.zeros(2))
+
+
+def test_generators_original_issue(jaxtyp, typecheck):
+    # Effectively the same as https://github.com/patrick-kidger/jaxtyping/issues/91
+    if torch is None:
+        pytest.skip("torch is not available")
+
+    @jaxtyp(typecheck)
+    def g(x: Shaped[torch.Tensor, "*"]) -> Iterator[Shaped[torch.Tensor, "*"]]:
+        yield x
+
+    @jaxtyp(typecheck)
+    def f():
+        next(g(torch.zeros(1)))
+        next(g(torch.zeros(2)))
+
+    f()

test/test_import_hook.py

@@ -17,76 +17,105 @@
 # IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 # CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
+import importlib
+import importlib.metadata
+import pathlib
+import shutil
+import sys
+import tempfile
+
 import pytest
 
-from jaxtyping import install_import_hook
+import jaxtyping
 
 
-def test_import_hook_typeguard_old():
-    hook = install_import_hook(
-        "test.import_hook_tester_typeguard_old", ("typeguard", "typechecked")
-    )
-    with hook:
-        from . import import_hook_tester_typeguard_old  # noqa: F401
+_here = pathlib.Path(__file__).parent
 
 
-def test_import_hook_typeguard():
-    hook = install_import_hook(
-        "test.import_hook_tester_typeguard", "typeguard.typechecked"
-    )
-    with hook:
-        from . import import_hook_tester_typeguard  # noqa: F401
-
-
-def test_import_hook_beartype_old():
+try:
+    typeguard_version = importlib.metadata.version("typeguard")
+except Exception as e:
+    raise ImportError("Could not find typeguard version") from e
+else:
     try:
-        import beartype  # noqa: F401
-    except ImportError:
-        pytest.skip("Beartype not installed")
-    else:
-        hook = install_import_hook(
-            "test.import_hook_tester_beartype_old", ("beartype", "beartype")
-        )
-        with hook:
-            from . import import_hook_tester_beartype_old  # noqa: F401
-
-
-def test_import_hook_beartype():
-    try:
-        import beartype  # noqa: F401
-    except ImportError:
-        pytest.skip("Beartype not installed")
-    else:
-        hook = install_import_hook(
-            "test.import_hook_tester_beartype", "beartype.beartype"
-        )
-        with hook:
-            from . import import_hook_tester_beartype  # noqa: F401
-
-
-def test_import_hook_beartype_full():
-    try:
-        import beartype  # noqa: F401
-    except ImportError:
-        pytest.skip("Beartype not installed")
-    else:
-        bearchecker = "beartype.beartype(conf=beartype.BeartypeConf(strategy=beartype.BeartypeStrategy.On))"  # noqa: E501
-        hook = install_import_hook("test.import_hook_tester_beartype_full", bearchecker)
-        with hook:
-            from . import import_hook_tester_beartype_full  # noqa: F401
-
-
-def test_import_hook_transitive():
-    hook = install_import_hook(
-        "test.import_hook_tester_transitive", "typeguard.typechecked"
+        major, _, _ = typeguard_version.split(".")
+        major = int(major)
+    except Exception as e:
+        raise ImportError(
+            f"Unexpected typeguard version {typeguard_version}; not formatted as "
+            "`major.minor.patch`"
+        ) from e
+if major != 2:
+    raise ImportError(
+        "jaxtyping's tests required typeguard version 2. (Versions 3 and 4 are both "
+        "known to have bugs.)"
     )
-    with hook:
-        from . import import_hook_tester_transitive  # noqa: F401
 
 
-def test_import_hook_broken_checker():
-    hook = install_import_hook(
-        "test.import_hook_tester_broken_checker", "jaxtyping.does_not_exist"
-    )
-    with hook, pytest.raises(AttributeError):
-        from . import import_hook_tester_broken_checker  # noqa: F401
+assert not hasattr(jaxtyping, "_test_import_hook_counter")
+jaxtyping._test_import_hook_counter = 0
+
+
+@pytest.fixture(scope="module")
+def importhook_tempdir():
+    with tempfile.TemporaryDirectory() as dir:
+        sys.path.append(dir)
+        dir = pathlib.Path(dir)
+        shutil.copyfile(_here / "helpers.py", dir / "helpers.py")
+        yield dir
+
+
+def _test_import_hook(importhook_tempdir, typechecker):
+    counter = jaxtyping._test_import_hook_counter
+    stem = f"import_hook_tester{counter}"
+    shutil.copyfile(_here / "import_hook_tester.py", importhook_tempdir / f"{stem}.py")
+
+    importlib.invalidate_caches()
+    with jaxtyping.install_import_hook(stem, typechecker):
+        importlib.import_module(stem)
+    assert counter + 1 == jaxtyping._test_import_hook_counter
+
+
+# Tests start below...
+
+
+def test_import_hook_typeguard(importhook_tempdir, typeguard_or_skip):
+    _test_import_hook(importhook_tempdir, "typeguard.typechecked")
+
+
+def test_import_hook_beartype(importhook_tempdir, beartype_or_skip):
+    _test_import_hook(importhook_tempdir, "beartype.beartype")
+
+
+def test_import_hook_beartype_full(importhook_tempdir, beartype_or_skip):
+    bearchecker = "beartype.beartype(conf=beartype.BeartypeConf(strategy=beartype.BeartypeStrategy.On))"  # noqa: E501
+    _test_import_hook(importhook_tempdir, bearchecker)
+
+
+def test_import_hook_typeguard_old(importhook_tempdir, typeguard_or_skip):
+    _test_import_hook(importhook_tempdir, ("typeguard", "typechecked"))
+
+
+def test_import_hook_beartype_old(importhook_tempdir, beartype_or_skip):
+    _test_import_hook(importhook_tempdir, ("beartype", "beartype"))
+
+
+def test_import_hook_broken_checker(importhook_tempdir):
+    with pytest.raises(AttributeError):
+        _test_import_hook(importhook_tempdir, "jaxtyping.does_not_exist")
+
+
+def test_import_hook_transitive(importhook_tempdir, typeguard_or_skip):
+    counter = jaxtyping._test_import_hook_counter
+    transitive_name = "jaxtyping_transitive_test"
+    transitive_dir = importhook_tempdir / transitive_name
+    transitive_dir.mkdir()
+    shutil.copyfile(_here / "import_hook_tester.py", transitive_dir / "tester.py")
+    with open(transitive_dir / "__init__.py", "w") as f:
+        f.write("from . import tester")
+        f.flush()
+
+    importlib.invalidate_caches()
+    with jaxtyping.install_import_hook(transitive_name, "typeguard.typechecked"):
+        importlib.import_module(transitive_name)
+    assert counter + 1 == jaxtyping._test_import_hook_counter

test/test_ipython_extension.py

@@ -0,0 +1,151 @@
+import pytest
+from IPython.testing.globalipapp import start_ipython
+
+from .helpers import ParamError
+
+
+@pytest.fixture(scope="session")
+def session_ip():
+    yield start_ipython()
+
+
+@pytest.fixture(scope="function")
+def ip(session_ip):
+    session_ip.run_cell(raw_cell="import jaxtyping")
+    session_ip.run_line_magic(magic_name="load_ext", line="jaxtyping")
+    session_ip.run_line_magic(
+        magic_name="jaxtyping.typechecker", line="typeguard.typechecked"
+    )
+    yield session_ip
+
+
+def test_that_ipython_works(ip):
+    ip.run_cell(raw_cell="x = 1").raise_error()
+    assert ip.user_global_ns["x"] == 1
+
+
+def test_function_beartype(ip):
+    ip.run_cell(
+        raw_cell="""
+    def f(x: int):
+        pass
+                """
+    ).raise_error()
+    ip.run_cell(raw_cell="f(1)").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell='f("x")').raise_error()
+
+
+def test_function_none(ip):
+    ip.run_cell(
+        raw_cell="""
+    def f(a,b,c):
+        pass
+                """
+    ).raise_error()
+    ip.run_cell(raw_cell='f(1,2,"k")').raise_error()
+
+
+def test_function_jaxtyped(ip):
+    ip.run_cell(
+        raw_cell="""
+    from jaxtyping import Float, Array, Int
+    import jax
+
+    def g(x: Float[Array, "1"]):
+        return x + 1
+
+                """
+    ).raise_error()
+
+    ip.run_cell(raw_cell="g(jax.numpy.array([1.0]))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell="g(jax.numpy.array(1.0))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell="g(jax.numpy.array([1]))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell="g(jax.numpy.array([2, 3]))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell='g("string")').raise_error()
+
+
+def test_function_jaxtyped_and_jitted(ip):
+    ip.run_cell(
+        raw_cell="""
+    from jaxtyping import Float, Array, Int
+    import jax
+
+    @jax.jit
+    def g(x: Float[Array, "1"]):
+        return x + 1
+
+                """
+    ).raise_error()
+
+    ip.run_cell(raw_cell="g(jax.numpy.array([1.0]))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell="g(jax.numpy.array(1.0))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell="g(jax.numpy.array([1]))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell="g(jax.numpy.array([2, 3]))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell='g("string")').raise_error()
+
+
+def test_class_jaxtyped(ip):
+    ip.run_cell(
+        raw_cell="""
+    from jaxtyping import Float, Array, Int
+    import equinox as eqx
+    import jax
+
+    class A(eqx.Module):
+        x: Float[Array, "2"]
+
+        def do_something(self, y: Int[Array, ""]):
+            return self.x + y
+                """
+    ).raise_error()
+
+    ip.run_cell(raw_cell="a = A(jax.numpy.array([1.0, 2.0]))").raise_error()
+    ip.run_cell(raw_cell="a.do_something(jax.numpy.array(2))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(raw_cell="A(jax.numpy.array([1.0]))").raise_error()
+
+    with pytest.raises(ParamError):
+        ip.run_cell(
+            raw_cell="a.do_something(jax.numpy.array([2.0, 3.0]))"
+        ).raise_error()
+
+
+def test_class_not_dataclass(ip):
+    ip.run_cell(
+        raw_cell="""
+    from jaxtyping import Float, Array, Int
+    import equinox as eqx
+    import jax
+
+    class A:
+        def __init__(self, x):
+            self.x = x
+
+        def do_something(self, y):
+            return x + y
+                """
+    ).raise_error()
+
+    ip.run_cell(raw_cell="a = A(jax.numpy.array([1.0, 2.0]))").raise_error()
+    ip.run_cell(raw_cell="a.do_something(jax.numpy.array(2))").raise_error()
+    ip.run_cell(raw_cell="A(jax.numpy.array([1.0]))").raise_error()
+    ip.run_cell(raw_cell="a.do_something(jax.numpy.array([2.0, 3.0]))").raise_error()

test/test_messages.py

@@ -0,0 +1,104 @@
+from typing import Any
+
+import equinox as eqx
+import jax.numpy as jnp
+import pytest
+
+from jaxtyping import Array, Float, jaxtyped, PyTree, TypeCheckError
+
+
+def test_arg_localisation(typecheck):
+    @jaxtyped(typechecker=typecheck)
+    def f(x: str, y: str, z: int):
+        pass
+
+    matches = [
+        "Type-check error whilst checking the parameters of f",
+        "The problem arose whilst typechecking parameter 'z'.",
+        "Called with parameters: {'x': 'hi', 'y': 'bye', 'z': 'not-an-int'}",
+        r"Parameter annotations: \(x: str, y: str, z: int\).",
+    ]
+    for match in matches:
+        with pytest.raises(TypeCheckError, match=match):
+            f("hi", "bye", "not-an-int")
+
+    @jaxtyped(typechecker=typecheck)
+    def g(x: Float[Array, "a b"], y: Float[Array, "b c"]):
+        pass
+
+    x = jnp.zeros((2, 3))
+    y = jnp.zeros((4, 3))
+    matches = [
+        "Type-check error whilst checking the parameters of g",
+        "The problem arose whilst typechecking parameter 'y'.",
+        r"Called with parameters: {'x': f32\[2,3\], 'y': f32\[4,3\]}",
+        (
+            r"Parameter annotations: \(x: Float\[Array, 'a b'\], y: "
+            r"Float\[Array, 'b c'\]\)."
+        ),
+        "The current values for each jaxtyping axis annotation are as follows.",
+        "a=2",
+        "b=3",
+    ]
+    for match in matches:
+        with pytest.raises(TypeCheckError, match=match):
+            g(x, y=y)
+
+
+def test_return(typecheck):
+    @jaxtyped(typechecker=typecheck)
+    def f(x: PyTree[Any, " T"], y: PyTree[Any, " S"]) -> PyTree[Any, "T S"]:
+        return "foo"
+
+    x = (1, 2)
+    y = {"a": 1}
+    matches = [
+        "Type-check error whilst checking the return value of f",
+        r"Called with parameters: {'x': \(1, 2\), 'y': {'a': 1}}",
+        "Actual value: 'foo'",
+        r"Expected type: PyTree\[Any, \"T S\"\].",
+        (
+            "The current values for each jaxtyping PyTree structure annotation are as "
+            "follows."
+        ),
+        r"T=PyTreeDef\(\(\*, \*\)\)",
+        r"S=PyTreeDef\({'a': \*}\)",
+    ]
+    for match in matches:
+        with pytest.raises(TypeCheckError, match=match):
+            f(x, y=y)
+
+
+def test_dataclass_attribute(typecheck):
+    @jaxtyped(typechecker=typecheck)
+    class M(eqx.Module):
+        x: Float[Array, " *foo"]
+        y: PyTree[Any, " T"]
+        z: int
+
+    x = jnp.zeros((2, 3))
+    y = (1, (3, 4))
+    z = "not-an-int"
+
+    matches = [
+        "Type-check error whilst checking the parameters of M",
+        "The problem arose whilst typechecking parameter 'z'.",
+        (
+            r"Called with parameters: {'self': M\(\.\.\.\), 'x': f32\[2,3\], "
+            r"'y': \(1, \(3, 4\)\), 'z': 'not-an-int'}"
+        ),
+        (
+            r"Parameter annotations: \(self: Any, x: Float\[Array, '\*foo'\], "
+            r"y: PyTree\[Any, \"T\"\], z: int\)."
+        ),
+        "The current values for each jaxtyping axis annotation are as follows.",
+        r"foo=\(2, 3\)",
+        (
+            "The current values for each jaxtyping PyTree structure annotation are as "
+            "follows."
+        ),
+        r"T=PyTreeDef\(\(\*, \(\*, \*\)\)\)",
+    ]
+    for match in matches:
+        with pytest.raises(TypeCheckError, match=match):
+            M(x, y, z)

test/test_no_jax_dependency.py

@@ -0,0 +1,25 @@
+import subprocess
+import sys
+
+
+_py_path = sys.executable
+
+
+def test_no_jax_dependency():
+    result = subprocess.run(
+        f"{_py_path} -c "
+        "'import jaxtyping; import sys; sys.exit(\"jax\" in sys.modules)'",
+        shell=True,
+    )
+    assert result.returncode == 0
+
+
+# Meta-test: test that the above test will work. (i.e. that I haven't messed up using
+# subprocess.)
+def test_meta():
+    result = subprocess.run(
+        f"{_py_path} -c 'import jaxtyping; import jax; import sys; "
+        'sys.exit("jax" in sys.modules)\'',
+        shell=True,
+    )
+    assert result.returncode == 1

test/test_pytree.py

@@ -25,7 +25,8 @@ import jax.numpy as jnp
 import jax.random as jr
 import pytest
 
-from jaxtyping import Float, jaxtyped, PyTree
+import jaxtyping
+from jaxtyping import AnnotationError, Array, Float, PyTree
 
 from .helpers import make_mlp, ParamError
 
@@ -92,9 +93,8 @@ def test_nested_pytrees(getkey, typecheck):
         g([1, 2, make_mlp()])
 
 
-def test_pytree_array(typecheck):
-    @jaxtyped
-    @typecheck
+def test_pytree_array(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
     def g(x: PyTree[Float[jnp.ndarray, "..."]]):
         pass
 
@@ -106,9 +106,8 @@ def test_pytree_array(typecheck):
         g(1.0)
 
 
-def test_pytree_shaped_array(typecheck, getkey):
-    @jaxtyped
-    @typecheck
+def test_pytree_shaped_array(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
     def g(x: PyTree[Float[jnp.ndarray, "b c"]]):
         pass
 
@@ -193,3 +192,153 @@ def test_subclass_pytree():
     assert issubclass(x, PyTree)
     assert issubclass(y, PyTree)
     assert not issubclass(int, PyTree)
+
+
+def test_structure_match(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f(x: PyTree[int, " T"], y: PyTree[str, " T"]):
+        pass
+
+    f(1, "hi")
+    f((3, 4, {"a": 5}), ("a", "b", {"a": "c"}))
+
+    with pytest.raises(ParamError):
+        f(1, ("hi",))
+
+
+def test_structure_prefix(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f(x: PyTree[int, " T"], y: PyTree[str, "T ..."]):
+        pass
+
+    f(1, "hi")
+    f((3, 4, {"a": 5}), ("a", "b", {"a": "c"}))
+    f(1, ("hi",))
+    f((1, 2), ({"a": "hi"}, {"a": "bye"}))
+    f((1, 2), ({"a": "hi"}, {"not-a": "bye"}))
+
+    with pytest.raises(ParamError):
+        f((1, 2), ({"a": "hi"}, {"a": "bye"}, {"a": "oh-no"}))
+
+    with pytest.raises(ParamError):
+        f((3, 4, 5), {"a": ("hi", "bye")})
+
+
+def test_structure_suffix(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f(x: PyTree[int, " T"], y: PyTree[str, "... T"]):
+        pass
+
+    f(1, "hi")
+    f((3, 4, {"a": 5}), ("a", "b", {"a": "c"}))
+    f(1, ("hi",))
+
+    with pytest.raises(ParamError):
+        f((3, 4), {"a": (1, 2)})
+
+    with pytest.raises(ParamError):
+        f((3, 4, 5), {"a": ("hi", "bye")})
+
+
+def test_structure_compose(jaxtyp, typecheck):
+    @jaxtyp(typecheck)
+    def f(x: PyTree[int, " T"], y: PyTree[int, " S"], z: PyTree[str, "S T"]):
+        pass
+
+    f(1, 2, "hi")
+    f((1, 2), 2, ("a", "b"))
+
+    with pytest.raises(ParamError):
+        f((1, 2), 2, (1, 2))
+
+    f((1, 2), {"a": 3}, {"a": ("hi", "bye")})
+
+    with pytest.raises(ParamError):
+        f((1, 2), {"a": 3}, ({"a": "hi"}, {"a": "bye"}))
+
+    @jaxtyp(typecheck)
+    def g(x: PyTree[int, " T"], y: PyTree[int, " S"], z: PyTree[str, "T S"]):
+        pass
+
+    with pytest.raises(ParamError):
+        g((1, 2), {"a": 3}, {"a": ("hi", "bye")})
+
+    g((1, 2), {"a": 3}, ({"a": "hi"}, {"a": "bye"}))
+
+
+@pytest.mark.parametrize("variadic", (False, True))
+def test_treepath_dependence_function(variadic, jaxtyp, typecheck, getkey):
+    if variadic:
+        jtshape = "*?foo"
+        shape = (2, 3)
+    else:
+        jtshape = "?foo"
+        shape = (4,)
+
+    @jaxtyp(typecheck)
+    def f(
+        x: PyTree[Float[Array, jtshape], " T"], y: PyTree[Float[Array, jtshape], " T"]
+    ):
+        pass
+
+    x1 = jr.normal(getkey(), shape)
+    y1 = jr.normal(getkey(), shape)
+    x2 = jr.normal(getkey(), (5,))
+    y2 = jr.normal(getkey(), (5,))
+    f(x1, y1)
+    f((x1, x2), (y1, y2))
+
+    with pytest.raises(ParamError):
+        f(x1, y2)
+
+    with pytest.raises(ParamError):
+        f((x1, x2), (y2, y1))
+
+
+@pytest.mark.parametrize("variadic", (False, True))
+def test_treepath_dependence_dataclass(variadic, typecheck, getkey):
+    if variadic:
+        jtshape = "*?foo"
+        shape = (2, 3)
+    else:
+        jtshape = "?foo"
+        shape = (4,)
+
+    @jaxtyping.jaxtyped(typechecker=typecheck)
+    class A(eqx.Module):
+        x: PyTree[Float[Array, jtshape], " T"]
+        y: PyTree[Float[Array, jtshape], " T"]
+
+    x1 = jr.normal(getkey(), shape)
+    y1 = jr.normal(getkey(), shape)
+    x2 = jr.normal(getkey(), (5,))
+    y2 = jr.normal(getkey(), (5,))
+    A(x1, y1)
+    A((x1, x2), (y1, y2))
+
+    with pytest.raises(ParamError):
+        A(x1, y2)
+
+    with pytest.raises(ParamError):
+        A((x1, x2), (y2, y1))
+
+
+def test_treepath_dependence_missing_structure_annotation(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
+    def f(x: PyTree[Float[Array, "?foo"], " T"], y: PyTree[Float[Array, "?foo"]]):
+        pass
+
+    x1 = jr.normal(getkey(), (2,))
+    y1 = jr.normal(getkey(), (2,))
+    with pytest.raises(AnnotationError, match="except when contained with structured"):
+        f(x1, y1)
+
+
+def test_treepath_dependence_multiple_structure_annotation(jaxtyp, typecheck, getkey):
+    @jaxtyp(typecheck)
+    def f(x: PyTree[PyTree[Float[Array, "?foo"], " S"], " T"]):
+        pass
+
+    x1 = jr.normal(getkey(), (2,))
+    with pytest.raises(AnnotationError, match="ambiguous which PyTree"):
+        f(x1)

test/test_serialisation.py

@@ -1,16 +1,25 @@
 import cloudpickle
 import numpy as np
-import torch
+
+
+try:
+    import torch
+except ImportError:
+    torch = None
 
 from jaxtyping import AbstractArray, Array, Shaped
 
 
 def test_pickle():
     x = cloudpickle.dumps(Shaped[Array, ""])
-    y = cloudpickle.dumps(AbstractArray)
-    z = cloudpickle.dumps(Shaped[np.ndarray, ""])
-    w = cloudpickle.dumps(Shaped[torch.Tensor, ""])
     cloudpickle.loads(x)
+
+    y = cloudpickle.dumps(AbstractArray)
     cloudpickle.loads(y)
+
+    z = cloudpickle.dumps(Shaped[np.ndarray, ""])
     cloudpickle.loads(z)
-    cloudpickle.loads(w)
+
+    if torch is not None:
+        w = cloudpickle.dumps(Shaped[torch.Tensor, ""])
+        cloudpickle.loads(w)

test/test_tf_dtype.py

@@ -0,0 +1,13 @@
+# Tensorflow dependency kept in a separate file, so that we can optionally exclude it
+# more easily.
+import tensorflow as tf
+
+from jaxtyping import UInt
+
+
+def test_tf_dtype():
+    x = tf.constant(1, dtype=tf.uint8)
+    y = tf.constant(1, dtype=tf.float32)
+    hint = UInt[tf.Tensor, "..."]
+    assert isinstance(x, hint)
+    assert not isinstance(y, hint)

test/test_threading.py

@@ -39,8 +39,7 @@ class _ErrorableThread(threading.Thread):
 
 
 def test_threading_jaxtyped():
-    @jaxtyped
-    @typechecked
+    @jaxtyped(typechecker=typechecked)
     def add(x: Float[Array, "a b"], y: Float[Array, "a b"]) -> Float[Array, "a b"]:
         return x + y