Re: [PATCH v7 05/11] osdep: build with non-working system() function

[Joelle van Dyne]

Here's how meson does cc.has_function

https://github.com/mesonbuild/meson/blob/master/mesonbuild/compilers/mixins/clike.py#L761

Since the compiler error comes from the header file with

__attribute__((availability(ios,unavailable)))

The meson check will always pass.

cc.compiles should work though. Is there a reason why it's not used
instead of all the compile_prog checks in ./configure ?

-j

On Mon, Jan 25, 2021 at 12:04 AM Paolo Bonzini <pbonzini@redhat.com> wrote:
>
> On 23/01/21 14:45, Peter Maydell wrote:
> > On Sat, 23 Jan 2021 at 03:18, Joelle van Dyne <j@getutm.app> wrote:
> > On Fri, Jan 22, 2021 at 3:17 PM Peter Maydell <peter.maydell@linaro.org> 
> > wrote:
> >>> Can we do the "does system() exist?" check in meson.build ?
> >
> >>> config_host_data.set('HAVE_SYSTEM_FUNCTION', cc.has_function('system'))
> >
> >> Unfortunately, this doesn't work for iOS, which defines system() but
> >> throws a compile time error if you try to call it.
> >
> > That's odd -- as far as I can tell the meson implementation
> > of has_function() does what I expected it to do, ie
> > "try to compile and link a little program that uses the
> > function, and see if it successfully links":
> > https://github.com/mesonbuild/meson/blob/39ede12aa5b27376341df85bc9ec254913f044bd/mesonbuild/compilers/mixins/clike.py#L791
> > There's some initial cleverness there too, so I guess some
> > part of that must be what's tripping us up.
> >
> > In any case, I think we should be doing new checks in
> > meson.build, not configure.  Paolo, what's the right
> > way to do a meson "really compile this program and
> > check it built" test?
>
> One possibility is that you have to specify the #include in the "prefix"
> argument of cc.has_function for the test to behave as the QEMU code?
>
> If cc.has_function doesn't work, there's cc.compiles() and cc.links().
>
> Paolo
>