Work around GCC -Warray-bounds false positive in argument_vector

[rwgk]

Description

Closescontourpy/contourpy#512

Validations:

• Build failure with master branch of pybind11 contourpy/contourpy#512 (comment)
• Build failure with master branch of pybind11 contourpy/contourpy#512 (comment)

Suggested changelog entry:

• A workaround for a GCC -Warray-bounds false positive in argument_vector was added.

[rwgk]

@swolchok does this look good to you? (Could you please approve?)

[swolchok]

@swolchok does this look good to you? (Could you please approve?)

Don’t you need to gate the push and the pop with the same conditions?

[rwgk]

Don’t you need to gate the push and the pop with the same conditions?

It's critical that the push/pop pair are balanced, but they don't have to be guarded. There are existing similar cases, for example here:

• pybind11/include/pybind11/cast.h

  Lines 1946 to 1951 in 734c29b

  	PYBIND11_WARNING_PUSH
  	#ifdef PYBIND11_DETECTED_CLANG_WITH_MISLEADING_CALL_STD_MOVE_EXPLICITLY_WARNING
  	PYBIND11_WARNING_DISABLE_CLANG("-Wreturn-std-move")
  	#endif
  	return result;
  	PYBIND11_WARNING_POP

The PYBIND11_WARNING_* macro system is designed so that we don't have to repeat complex #ifs. An extra PYBIND11_WARNING_PUSH/POP when they don't end up disabling anything is basically a no-op at compile time. It's just one more diagnostic push/pop pragma for the compiler to parse, which I believe is immeasurable compared to the cost of processing the surrounding templates.

[swolchok]

Seems fine then

[rwgk]

Thanks @swolchok!