Re: [PATCH] meson: fix botched compile check conversions

[Daniel P . Berrangé]

On Tue, Nov 16, 2021 at 11:51:16AM +0000, Peter Maydell wrote:
> On Tue, 16 Nov 2021 at 09:38, Paolo Bonzini <pbonzini@redhat.com> wrote:
> >
> > Fix a bunch of incorrect conversions from configure to Meson, which result
> > in different outcomes with --extra-cflags=-Werror.
> 
> FWIW this still won't give the right answer for the 'struct iovec'
> test if you include -Werror via --extra-cflags, because the
> generated code trips over an "expression result unused" warning:
> 
> 
> Running compile:
> Working directory:
> /mnt/nvmedisk/linaro/qemu-from-laptop/qemu/build/arm-clang/meson-private/tmpeiz36t2n
> Command line:  clang-7 -m64 -mcx16
> /mnt/nvmedisk/linaro/qemu-from-laptop/qemu/build/arm-clang/meson-private/tmpeiz36t2n/testfile.c
> -o /mnt/nvmedis
> k/linaro/qemu-from-laptop/qemu/build/arm-clang/meson-private/tmpeiz36t2n/output.obj
> -c -fsanitize=undefined -fno-sanitize=shift-base -Werror -D_FI
> LE_OFFSET_BITS=64 -O0 -Werror=implicit-function-declaration
> -Werror=unknown-warning-option -Werror=unused-command-line-argument
> -Werror=ignored-op
> timization-argument -std=gnu11
> 
> Code:
>  #include <sys/uio.h>
>         void bar(void) {
>             sizeof(struct iovec);
>         };
> Compiler stdout:
> 
> Compiler stderr:
>  
> /mnt/nvmedisk/linaro/qemu-from-laptop/qemu/build/arm-clang/meson-private/tmpeiz36t2n/testfile.c:3:13:
> error: expression result unused [-Werror,-Wunused-value]
>             sizeof(struct iovec);
>             ^~~~~~~~~~~~~~~~~~~~
> 1 error generated.
> 
> Checking for type "struct iovec" : NO
> 
> 
> But maybe we should just explicitly reject -Werror in --extra-cflags...

I wonder if the problem is more fundamental than that. Passing
stuff in --extra-cflags is done to influence the flags used to
compile the QEMU end user binaries. Unfortunately --extra-cflags
is also getting applied to all the meson.build feature checks.

IMHO we would get a more reliable result if the meson.build
checks were fully isolated from the cflags we used for building
everything else, so the checks get a well understood, predictable
environment.

IIUC, this current behaviour is a result of us adding cflags
using  add_global_arguments / add_project_arguments.

I wonder if we need to exclusively use the 'c_args' parameter
to library()/executable() and friends ?  The downside is of
course would be extra work to make sure we pass c_args in all
the right places.


Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|