Use defined for PYBIND11_HAS_OPTIONAL, PYBIND11_HAS_EXP_OPTIONAL, PYBIND11_HAS_VARIANT

[cirosantilli2]

These variables are not set in certain circumstances, and if the project
using pybind11 sets -Wundef, the warnings will show.

In all other usages outside of stl.h, ifdef/defined has already been used,
this commit converts the remaining ones under stl.h.

The test build is also modified to catch the problem.

The errors looked like:

ext/pybind11/include/pybind11/stl.h:292:5: error: "PYBIND11_HAS_OPTIONAL" is not defined, evaluates to 0 [-Werror=undef]
292 | #if PYBIND11_HAS_OPTIONAL
| ^~~~~~~~~~~~~~~~~~~~~
ext/pybind11/include/pybind11/stl.h:300:5: error: "PYBIND11_HAS_EXP_OPTIONAL" is not defined, evaluates to 0 [-Werror=undef]
300 | #if PYBIND11_HAS_EXP_OPTIONAL
| ^~~~~~~~~~~~~~~~~~~~~~~~~
ext/pybind11/include/pybind11/stl.h:372:5: error: "PYBIND11_HAS_VARIANT" is not defined, evaluates to 0 [-Werror=undef]
372 | #if PYBIND11_HAS_VARIANT
| ^~~~~~~~~~~~~~~~~~~~

[henryiii]

Note this changes the behavior; defining this to 0 causes it to be used now, instead of skipped. A completely backward compatible version would be #if defined(PYBIND11_HAS_OPTIONAL) && PYBIND11_HAS_OPTIONAL.

[YannickJadoul]

Note this changes the behavior; defining this to 0 causes it to be used now, instead of skipped. A completely backward compatible version would be #if defined(PYBIND11_HAS_OPTIONAL) && PYBIND11_HAS_OPTIONAL.

We're detecting and defining this ourselves, now, so... I don't think this is documented/meant to be set from the outside?

[henryiii]

We're detecting and defining this ourselves, now, so...

Probably okay, then. We need to change several #if (__APPLE__)'s to #if defined(__APPLE__) and such.

[YannickJadoul]

(Just wait for @bstaletic to pop up and shout about Hyrum's law, though.)

But another argument: pybind11's code detecting it also doesn't check if something is already defined:

pybind11/include/pybind11/stl.h

Lines 27 to 49 in fbc7563

	#ifdef __has_include
	// std::optional (but including it in c++14 mode isn't allowed)
	# if defined(PYBIND11_CPP17) && __has_include(<optional>)
	# include<optional>
	# definePYBIND11_HAS_OPTIONAL1
	# endif
	// std::experimental::optional (but not allowed in c++11 mode)
	# if defined(PYBIND11_CPP14) && (__has_include(<experimental/optional>) && \
	!__has_include(<optional>))
	# include<experimental/optional>
	# definePYBIND11_HAS_EXP_OPTIONAL1
	# endif
	// std::variant
	# if defined(PYBIND11_CPP17) && __has_include(<variant>)
	# include<variant>
	# definePYBIND11_HAS_VARIANT1
	# endif
	#elif defined(_MSC_VER) && defined(PYBIND11_CPP17)
	# include<optional>
	# include<variant>
	# definePYBIND11_HAS_OPTIONAL1
	# definePYBIND11_HAS_VARIANT1
	#endif

So if someone manually defines it, it's 1) overwrtiten (I think ?), and 2) likely gives some warning. So even considering a mild version of Hyrum's law, we should be fine? :-)

Would you mind having a look for the other macros that @henryiii mentions, and adding them to this PR, @cirosantilli2? :-)

[cirosantilli2]

I've updated the PR to update __APPLE__ and __clang__ too, there however only one instance of those not using defined. Let me know if you can spot any others.

[henryiii]

Your warning addition should spot them 👍

[henryiii]

/__w/pybind11/pybind11/tests/test_stl.cpp:18:5: error: 'PYBIND11_HAS_VARIANT' is not defined, evaluates to 0 [-Werror,-Wundef]
57
#if PYBIND11_HAS_VARIANT

[YannickJadoul]

@cirosantilli2 If I'm not mistaken, warnings are treated as errors in the CI builds. So just follow the smell of the warnings/errors ;-)