Re: [PATCH 05/12] compiler.h: drop __printf__ macro MinGW/glib workaround

[Peter Maydell]

On Wed, 16 Mar 2022 at 13:44, Daniel P. Berrangé <berrange@redhat.com> wrote:
>
> On Thu, Feb 24, 2022 at 08:14:47PM +0000, Peter Maydell wrote:
> > On Thu, 24 Feb 2022 at 19:50, Marc-André Lureau
> > <marcandre.lureau@gmail.com> wrote:
> > > On Thu, Feb 24, 2022 at 11:23 PM Peter Maydell <peter.maydell@linaro.org> 
> > > wrote:
> > >> You're probably building with a newer glib, and possibly also
> > >> a newer mingw.
> > >>
> > >> I've cc'd Stefan Weil who might know whether we can drop this
> > >> workaround as far as the mingw part is concerned.
> > >
> > >
> > > Probably safer to keep it until we bump glib dependency to >=2.58.
> > >
> > > I would move it to glib-compat.h though, and leave a note there, as it is 
> > > (or should be ) an old glib specific workaround.
> >
> > We can only move it to glib-compat if we confirm that only the
> > glib-related part of the workaround is still relevant and the
> > mingw side is now no longer needed, though.
>
> We know glib uses the GNU printf semantics for all its APIs.
>
> We know QEMU code will use the GNU printf annotation for all its
> APIs where it knows it has GNU printf, due to delegating to
> GLib.
>
> For 3rd party libraries, we can have no confidence about whether
> they expect GNU or native printf format, unless we're doing
> something to override the printf family of functions at link
> time. IIRC, we're not doing that, so we can't assume 3rd party
> stuff expects GNU format, and so the sooner we get rid of
> the #define __printf__ __gnu_printf__ the better IMHO. The
> proof of course would be to see a CI test run with the define
> removed proving that no code we call relies on it.

Yes, the workaround is definitely correct for QEMU's own
code and for glib itself. We don't care about 3rd party
libraries because we don't use any of those which take
format-string arguments AFAIK. The 'mingw' part AIUI is
purely for mingw itself, ie the standard library. What
I'm asking is "what were the versions of mingw that were
affected by this, and are they all old enough we don't need
to care from that point of view?".

-- PMM