Wrap va_list function as variadic function

[Urgau]

Opening a draft before doing more to get opinion on it.

This PR add a way to wrap va_list function as variadic function. See below for an example.

Input:

#include <stdarg.h>

static inline int vfoo(int flags, va_list va) {
    // ...
    return flags;
}

Output:

int vfoo__extern(int flags, ...) {
int ret;
va_list ap;

va_start(ap, flags);
ret = vfoo(flags, ap);
va_end(ap);
return ret;
}

extern "C" {
    #[link_name = "vfoo__extern"]
    pub fn vfoo(flags: ::std::os::raw::c_int, ...) -> ::std::os::raw::c_int;
}

Related to #2498

cc @pvdrz

[bors-servo]

☔ The latest upstream changes (presumably 4f9fcbf) made this pull request unmergeable. Please resolve the merge conflicts.

[Urgau]

@pvdrz Would you be able to give me pre-review?

[pvdrz]

I think it would be best if this is not done for every single function that takes valist as a last argument but instead let it be configurable using ParseCallbacks:

trait ParseCallbacks {
    ...
    fn wrap_as_variadic_fn(&mut self, name: &str) -> Option<String> {
        None
    }
}

That would mean that the user can decide which functions should be wrapped and would be their responsibility to check the signature instead of us.

For example, if callbacks.wrap_as_variadic_fn("vfoo") returns Some("foo"), we emit the following code:

int foo(int flags, ...) {
int ret;
va_list ap;

va_start(ap, flags);
ret = vfoo(flags, ap);
va_end(ap);
return ret;
}

extern "C" {
    pub fn foo(flags: ::std::os::raw::c_int, ...) -> ::std::os::raw::c_int;
}

What do you think?

[Urgau]

I think it would be best if this is not done for every single function that takes valist as a last argument but instead let it be configurable using ParseCallbacks:

[...]

What do you think?

Sure, this seems like a good idea. I will do the necessary work and make the PR ready for review.

Btw, my logic works with any function that has 1 (and only 1) va_list no matter his position. (I went trough great length to ensure that)

[Urgau]

@pvdrz As discussed I've added ParseCallbacks::wrap_as_variadic_fn. I also did some cleanup and improvement to the added code. Should be ready to review.

Btw, I ended up only changing the name of the Rust function because changing the other one felt unnecessary complicated and also it's the only one that matters (the other one are implementation details).

[Urgau]

@pvdrz Let me know if there is something I can do to help with the review ;-)

[pvdrz]

@pvdrz Let me know if there is something I can do to help with the review ;-)

Hey :) I have this on my todo list but I only will be able to review it later this week. Thanks for the patience

[Urgau]

@pvdrz The PR is ready for review.

(btw, do you want pings from me?)

[pvdrz]

This looks good to me. The only two extra things I'd ask you to do are:

• Check how wrap-static-fns is tested in bindgen-integration and update the tests to include this new feature.
• Add the changes to the changelog.

[Urgau]

@pvdrz I've updated as requested the PR (bindgen-integration + changelog).

[pvdrz]

Hey! the test suite broke due to an unrelated issue that has already been fixed in the main branch, could you rebase your PR against it so we can be sure the tests are passing?

[Urgau]

I've rebased, the tests suite now passes (had to add an exception for stdarg.h which I don't think is an issue, it's avalaible since C89 and even a little before) ;-)

[bors-servo]

☔ The latest upstream changes (presumably 4faa366) made this pull request unmergeable. Please resolve the merge conflicts.