[FFmpeg-devel] [PATCHv3] avutil/attributes: extend av_uninit to clang

[Ronald S. Bultje]

Hi,

On Wed, Oct 7, 2015 at 8:50 PM, Ganesh Ajjanagadde <gajjanag at mit.edu> wrote:

> On Wed, Oct 7, 2015 at 8:21 PM, Michael Niedermayer
> <michael at niedermayer.cc> wrote:
> > On Wed, Oct 07, 2015 at 06:22:08PM -0400, Ganesh Ajjanagadde wrote:
> >> Commit 6dac8c8327 disabled av_uninit for clang, due to some useless
> warnings.
> >> The concept of av_uninit is inherently useful though. This patch
> silences a
> >> bunch of warnings on clang e.g
> >>
> http://fate.ffmpeg.org/log.cgi?time=20150918181527&log=compile&slot=x86_64-darwin-clang-polly-vectorize-stripmine-3.7
> .
> >> Furthermore, it should be useful for general usage of av_uninit in
> future.
> >>
> >> Signed-off-by: Ganesh Ajjanagadde <gajjanagadde at gmail.com>
> >> ---
> >>  libavutil/attributes.h | 8 ++++++++
> >>  1 file changed, 8 insertions(+)
> >>
> >> diff --git a/libavutil/attributes.h b/libavutil/attributes.h
> >> index 5c6b9de..ceb9e29 100644
> >> --- a/libavutil/attributes.h
> >> +++ b/libavutil/attributes.h
> >> @@ -147,6 +147,14 @@
> >>
> >>  #if defined(__GNUC__) && !defined(__INTEL_COMPILER) &&
> !defined(__clang__)
> >>  #    define av_uninit(x) x=x
> >> +#elif defined(__clang__)
> >> +#if __has_warning("-Wuninitialized")
> >> +#    define av_uninit(x) \
> >> +        _Pragma("clang diagnostic push") \
> >> +        _Pragma("clang diagnostic ignored \"-Wuninitialized\"") \
> >> +        x=x \
> >> +        _Pragma("clang diagnostic pop")
> >> +#endif
> >
> > this breaks
> > make distclean ; ./configure --cc='ccache clang' && make -j12
> > it does not break clang alone and clearing the ccache cache does not
> > help
> > ccache version 3.1.6
> >
> > In file included from libavfilter/af_aecho.c:1:
> > libavfilter/af_aecho.c:226:315: error: expected identifier or '('
> > static void echo_samples_dblp(AudioEchoContext *ctx, uint8_t
> **delayptrs, uint8_t * const *src, uint8_t **dst, int nb_samples, int
> channels) { const double out_gain = ctx->out_gain; const double in_gain =
> ctx->in_gain; const int nb_echoes = ctx->nb_echoes; const int max_samples =
> ctx->max_samples; int i, j, chan,#pragma clang diagnostic push#pragma clang
> diagnostic ignored "-Wuninitialized" index=index#pragma clang diagnostic
> pop; ((void)0); for (chan = 0; chan < channels; chan++) { const double *s =
> (double *)src[chan]; double *d = (double *)dst[chan]; double *dbuf =
> (double *)delayptrs[chan]; index = ctx->delay_index; for (i = 0; i <
> nb_samples; i++, s++, d++) { double out, in; in = *s; out = in * in_gain;
> for (j = 0; j < nb_echoes; j++) { int ix = index + max_samples -
> ctx->samples[j]; ix = (((ix) >= (max_samples)) ? (ix) - (max_samples) :
> (ix)); out += dbuf[ix] * ctx->decay[j]; } out *= out_gain; *d =
> av_clipd_c(out, -1.0, 1.0); dbuf[index] = in; index = (((index + 1) >=
> (max_samples)) ? (index + 1) - (max_samples) : (index + 1)); } }
> ctx->delay_index = index; }
> >
>
>
>
>              ^
> >
> > [...]
>
> ccache and clang have some weird interaction issues -for instance a
> lot of -Warray-bounds are triggered for glibc functions like strcmp. I
> vaguely recall some suggestion to use an environment variable to avoid
> the breakage:
> http://petereisentraut.blogspot.com/2011/09/ccache-and-clang-part-2.html
>
> Anyway, this sounds like it needs to be fixed upstream with clang or
> ccache or possibly both.


I agree, and am a little sad that the patch can't go in yet. Would you mind
filing a bug with ccache? According to that blog post, they (ccache devs)
do look at their bug tracker, and we have all information necessary to make
a relatively trivial test case for them (just the av_uninit define and a
one-line main() function that uses av_uninit should do the trick).

Ronald