Re: [RFC PATCH 4/4] ui/cocoa: Ignore 'initializer overrides' warnings

[Peter Maydell]

On Tue, 15 Feb 2022 at 13:18, Christian Schoenebeck
<qemu_oss@crudebyte.com> wrote:
>
> On Dienstag, 15. Februar 2022 13:06:25 CET Philippe Mathieu-Daudé via wrote:
> > We globally ignore the 'initializer overrides' warnings in C code
> > since commit c1556a812a ("configure: Disable (clang)
> > initializer-overrides warnings"). Unfortunately the ${gcc_flags}
> > variable is not propagated to Objective-C flags ($OBJCFLAGS).
> > Instead of reworking the configure script to test all supported
> > C flags in Objective-C and sanitize them, ignore the warning
> > locally with the GCC diagnostic #pragma (Clang ignores it).
> >
> > Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
> > ---
>
> What about this instead:
>
> diff --git a/ui/cocoa.m b/ui/cocoa.m
> index ac18e14ce0..df9b9be999 100644
> --- a/ui/cocoa.m
> +++ b/ui/cocoa.m
> @@ -652,9 +652,7 @@ QemuCocoaView *cocoaView;
>
>      /* translates Macintosh keycodes to QEMU's keysym */
>
> -    int without_control_translation[] = {
> -        [0 ... 0xff] = 0,   // invalid key
> -
> +    int without_control_translation[256] = {

I think this won't zero-initialize, because this is a function
level variable, not a global or static, but maybe ObjectiveC
rules differ from C here? (Besides, it really should be
a static const array.) That said...

> That warning should only be raised on overlapping designated initializations
> which strictly is undefined behaviour. Because which order defines the
> precedence of overlapping initializers, is it the order of the designated
> intializer list, or rather the memory order?

This is defined: textually last initializer wins.
https://gcc.gnu.org/onlinedocs/gcc/Designated-Inits.html

> See also:
> https://stackoverflow.com/questions/40920714/is-full-followed-by-partial-initialization-of-a-subobject-undefined-behavior

That's about struct initializers; we're dealing with array initializers here.

> So I have my doubts whether this warning suppression should be used in QEMU at
> all.

The warning is useful in the pure-standard-C world where there
is no such thing as the "range initialization" syntax. In that
case trying to initialize the same array member twice is very
likely a bug. However, if you use range initialization then
the pattern "initialize a range first, then override some specific
members within it" is very common and the warning is incorrect here.
We use and like the range-initialization syntax, and so we disable
the -Winitializer-overrides warnings. The bug here is just that
we are incorrectly failing to apply the warning flags we use
for C code when we compile ObjC files.

thanks
-- PMM