* ci: support Python 3.11-dev
Also update 3.10 to final, better PyPy usage
* fix: use PyFrame_GetCode on Python 3.9+
* ci: some bitiness of pypy not supported on win
* chore: update CMake support to 3.22rc1 to quiet warning
* fix: use dev version of py to fix Py 3.11
* tests: print proper Eigen version
* ci: include pypy2, not sure why
* ci: avoid running on Python 3.11 for now
* ci: fix runs
* ci: simpler PyPy usage, drop unmaintained scipy + pypy index
* ci: only binary numpy, wait on pypy 3.8
* refactor: address review
* fix: the types for return_value_policy_override in optional_caster
`return_value_policy_override` was not being applied correctly in
`optional_caster` in two ways:
- The `is_lvalue_reference` condition referenced `T`, which was the
`optional<T>` type parameter from the class, when it should have used `T_`,
which was the parameter to the `cast` function. `T_` can potentially be a
reference type, but `T` will never be.
- The type parameter passed to `return_value_policy_override` should be
`T::value_type`, not `T`. This matches the way that the other STL container
type casters work.
The result of these issues was that a method/property definition which used a
`reference` or `reference_internal` return value policy would create a Python
value that's bound by reference to a temporary C++ object, resulting in
undefined behavior. For reasons that I was not able to figure out fully, it
seems like this causes problems when using old versions of `boost::optional`,
but not with recent versions of `boost::optional` or the `libstdc++`
implementation of `std::optional`. The issue (that the override to
`return_value_policy::move` is never being applied) is present for all
implementations, it just seems like that somehow doesn't result in problems for
the some implementation of `optional`. This change includes a regression type
with a custom optional-like type which was able to reproduce the issue.
Part of the issue with using the wrong types may have stemmed from the type
variables `T` and `T_` having very similar names. This also changes the type
variables in `optional_caster` to use slightly more descriptive names, which
also more closely follow the naming convention used by the other STL casters.
Fixes#3330
* Fix clang-tidy complaints
* Add missing NOLINT
* Apply a couple more fixes
* fix: support GCC 4.8
* tests: avoid warning about unknown compiler for compilers missing C++17
* Remove unneeded test module attribute
* Change test enum to have more unique int values
Co-authored-by: Aaron Gokaslan <skylion.aaron@gmail.com>
Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>
* Adding MSVC C4127 suppression around Eigen includes.
* For MSVC 2015 only: also adding the C4127 suppression to test_eigen.cpp
* Copying original change from PR #3343, with extra line breaks to not run past 99 columns (our desired but currently not enforced limit).
* Add a test showing a flaw in make_key_iterator/make_value_iterator
If the iterator dereference operator returns a value rather than a
reference (and that pair also does not *contain* references),
make_key_iterator and make_value_iterator will return a reference to a
temporary, causing a segfault.
* Fix make_key_iterator/make_value_iterator for prvalue iterators
If an iterator returns a pair<T1, T2> rather than a reference to a pair
or a pair of references, make_key_iterator and make_value_iterator would
return a reference to a temporary, typically leading to a segfault. This
is because the value category of member access to a prvalue is an
xvalue, not a prvalue, so decltype produces an rvalue reference type.
Fix the type calculation to handle this case.
I also removed some decltype parentheses that weren't needed, either
because the expression isn't one of the special cases for decltype or
because decltype was only used for SFINAE. Hopefully that makes the code
a bit more readable.
Closes#3347
* Attempt a workaround for nvcc
* Add `.keys` and `.values` to bind_map
Both of these implement views (rather than just iterators), and `.items`
is also upgraded to a view. In practical terms, this allows a view to be
iterated multiple times and have its size taken, neither of which works
with an iterator.
The views implement `__len__`, `__iter__`, and the keys view implements
`__contains__`. Testing membership also works in item and value views
because Python falls back to iteration. This won't be optimal
for item values since it's linear rather than O(log n) or O(1), but I
didn't fancy trying to get all the corner cases to match Python
behaviour (tuple of wrong types, wrong length tuple, not a tuple etc).
Missing relative to Python dictionary views is `__reversed__` (only
added to Python in 3.8). Implementing that could break code that binds
custom map classes which don't provide `rbegin`/`rend` (at least without
doing clever things with SFINAE), so I've not tried.
The size increase on my system is 131072 bytes, which is rather large
(5%) but also suspiciously round (2^17) and makes me suspect some
quantisation effect.
* bind_map: support any object in __contains__
Add extra overload of `__contains__` (for both the map itself and
KeysView) which takes an arbitrary object and returns false.
* Take py::object by const reference in __contains__
To keep clang-tidy happy.
* Removing stray `py::` (detected via interactive testing in Google environment).
Co-authored-by: Ralf W. Grosse-Kunstleve <rwgk@google.com>
* Docs: Demonstrate non-enum internal types in example
Previously example only demonstrated internal enumeration type.
To show that it works for other internal types the same way the example was updated with an additional struct Pet::Attributes type.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Fix `pybind11::object::operator=` to be safe if `*this` is accessible from Python
* Add `custom_type_setup` attribute
This allows for custom modifications to the PyHeapTypeObject prior to
calling `PyType_Ready`. This may be used, for example, to define
`tp_traverse` and `tp_clear` functions.
The new FindPython-based variant of the CMake scripts caches information
about the chosen Python version that can become stale. For example,
suppose I configure a simple pybind11-based project as follows
```
cmake -S . -B build -GNinja -DPython_ROOT=<path to python 3.8>
```
which will generate `my_extension.cpython-38-x86_64-linux-gnu.so`.
A subsequent change to the python version like
```
cmake -S . -B build -GNinja -DPython_ROOT=<path to python 3.9>
```
does not update all necessary build system information. In particular,
the compiled file is still called
`my_extension.cpython-38-x86_64-linux-gnu.so`.
This commit fixes the problem by detecting changes in
`Python_EXECUTABLE` and re-running Python as needed.
Note that the previous way of detecting Python does not seem to be
affected, it always specifies the right suffix.
* Add make_value_iterator (#3271)
* Add make_value_iterator
This is the counterpart to make_key_iterator, and will allow
implementing a `value` method in `bind_map` (although doing so is left
for a subsequent PR).
I made a few design changes to reduce copy-and-paste boilerplate.
Previously detail::iterator_state had a boolean template parameter to
indicate whether it was being used for make_iterator or
make_key_iterator. I replaced the boolean with a class that determines
how to dereference the iterator. This allows for a generic
implementation of `__next__`.
I also added the ValueType and Extra... parameters to the iterator_state
template args, because I think it was a bug that they were missing: if
make_iterator is called twice with different values of these, only the
first set has effect (because the state class is only registered once).
There is still a potential issue in that the *values* of the extra
arguments are latched on the first call, but since most policies are
empty classes this should be even less common.
* Add some remove_cv_t to appease clang-tidy
* Make iterator_access and friends take reference
For some reason I'd accidentally made it take a const value, which
caused some issues with third-party packages.
* Another attempt to remove remove_cv_t from iterators
Some of the return types were const (non-reference) types because of the
pecularities of decltype: `decltype((*it).first)` is the *declared* type
of the member of the pair, rather than the type of the expression. So if
the reference type of the iterator is `pair<const int, int> &`, then the
decltype is `const int`. Wrapping an extra set of parentheses to form
`decltype(((*it).first))` would instead give `const int &`.
This means that the existing make_key_iterator actually returns by value
from `__next__`, rather than by reference. Since for mapping types, keys
are always const, this probably hasn't been noticed, but it will affect
make_value_iterator if the Python code tries to mutate the returned
objects. I've changed things to use double parentheses so that
make_iterator, make_key_iterator and make_value_iterator should now all
return the reference type of the iterator. I'll still need to add a test
for that; for now I'm just checking whether I can keep Clang-Tidy happy.
* Add back some NOLINTNEXTLINE to appease Clang-Tidy
This is favoured over using remove_cv_t because in some cases a const
value return type is deliberate (particularly for Eigen).
* Add a unit test for iterator referencing
Ensure that make_iterator, make_key_iterator and make_value_iterator
return references to the container elements, rather than copies. The
test for make_key_iterator fails to compile on master, which gives me
confidence that this branch has fixed it.
* Make the iterator_access etc operator() const
I'm actually a little surprised it compiled at all given that the
operator() is called on a temporary, but I don't claim to fully
understand all the different value types in C++11.
* Attempt to work around compiler bugs
https://godbolt.org/ shows an example where ICC gets the wrong result
for a decltype used as the default for a template argument, and CI also
showed problems with PGI. This is a shot in the dark to see if it fixes
things.
* Make a test constructor explicit (Clang-Tidy)
* Fix unit test on GCC 4.8.5
It seems to require the arguments to the std::pair constructor to be
implicitly convertible to the types in the pair, rather than just
requiring is_constructible.
* Remove DOXYGEN_SHOULD_SKIP_THIS guards
Now that a complex decltype expression has been replaced by a simpler
nested type, I'm hoping Doxygen will be able to build it without issues.
* Add comment to explain iterator_state template params
* fix: regression in #3271
Co-authored-by: Bruce Merry <1963944+bmerry@users.noreply.github.com>
