add_project_dependencies() ignores 'preserve' include_type property

[artem]

Describe the bug
If the dependency passed to add_project_dependencies has 'preserve' include_type property (default), Meson downgrades it to 'non-system'

To Reproduce

add_project_dependencies(
  declare_dependency(
    include_directories: include_directories('third_party/my_3p_dep', is_system: true),
  )
)

Related code: https://github.com/mesonbuild/meson/blob/6736a54/mesonbuild/interpreter/interpreter.py#L2961-L2966

Expected behavior
Meson produces -isystemthird_party/my_3p_dep cmdline argument instead of -Ithird_party/my_3p_dep

system parameters

• Meson 1.5.1

[eli-schwartz]

This is your daily reminder that -isystem is a flag for controlling the #include_next priority ordering etc. and only makes sense to use when you are implementing a libc or something along those lines -- "my_3p_dep" seems like a very bad and off-target thing to be using this for?

[artem]

only makes sense to use when you are implementing a libc

Actually not. -isystem is the only viable way to silence warnings coming from 3rd party headers. Silencing them in entire project is not an option, and fixing them is problematic, since you do not own the code you want to use. A real pain with C++ header-only libraries, I would say.

Anyway, it's an off-topic discussion that does not belittle the importance of this issue.

[eli-schwartz]

only makes sense to use when you are implementing a libc

Actually not. -isystem is the only viable way to silence warnings coming from 3rd party headers. Silencing warnings is not an option, and fixing them is problematic, since you do not own the code you want to use.

It is literally not documented to silence warnings, even as a side effect of the thing it is actually documented to do. You have no guarantee that it will or won't do any such thing.

In return for this, you break the C++ compilation model.

Anyway, it's an off-topic discussion that does not belittle the importance of this issue.

If there is no real world use case of this "feature" then that affects whether people are interested in working on it.

Patches welcome, I guess.

[artem]

It is literally not documented to silence warnings

I think, the documentation says for itself: https://gcc.gnu.org/onlinedocs/cpp/System-Headers.html

All warnings, other than those generated by ‘#warning’ (see Diagnostics), are suppressed while GCC is processing a system header.

In return for this, you break the C++ compilation model.

I have a problem understanding what compilation model you are referring to. Could you please elaborate for the sake of completeness of the discussion?

Patches welcome, I guess

If this is an acknowledgment of the bug, I am happy to prepare the patches.