"Ronald S. Bultje" <[email protected]> writes: > Hi, > > 2012/1/31 Måns Rullgård <[email protected]>: >> "Ronald S. Bultje" <[email protected]> writes: >>> 2012/1/31 Måns Rullgård <[email protected]>: >>>> "Ronald S. Bultje" <[email protected]> writes: >>>> >>>>> void *av_fast_realloc(void *ptr, unsigned int *size, size_t min_size) >>>>> { >>>>> if(min_size < *size) >>>>> @@ -793,7 +839,7 @@ int attribute_align_arg avcodec_open2(AVCodecContext >>>>> *avctx, AVCodec *codec, AVD >>>>> } >>>>> >>>>> if(avctx->codec->init && >>>>> !(avctx->active_thread_type&FF_THREAD_FRAME)){ >>>>> - ret = avctx->codec->init(avctx); >>>>> + ret = testxmmclobbers(avctx->codec->init, avctx); >>>>> if (ret < 0) { >>>>> goto free_and_end; >>>>> } >>>> >>>> You're kidding, right? >>> >>> It's not my code, this is really what Ramiro wrote. I'm open for >>> better solutions, I just want this test in so Michael Kostylev can add >>> this as a fate machine (h264 on Win64 is broken with this enabled >>> right now - I want to catch regressions once I've fixed it, and right >>> now we can't). >> >> I appreciate the good cause, but this is just too ugly. > > I agree the current code is ugly. Can we un-uglify it?
[...] > And we can think of other things also. Again, this is (IMO) critical > to prevent future regressions on Win64, I've just been tracking down a > couple that Michael Kostylev detected, and I'd prefer to not break > this again all too often in the near future. Does the --wrap linker flag [1] work on Windows? If it does, and I see no reason for it not to, this could be used to wrap avcodec_decode_foo() calls with checking functions in a separate file. [1] http://sourceware.org/binutils/docs-2.22/ld/Options.html#index-g_t_002d_002dwrap_003d_0040var_007bsymbol_007d-263 -- Måns Rullgård [email protected] _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
