[FFmpeg-devel-irc] IRC log for 2010-04-03

[irc at mansr.com]

[09:33:33] <merbanan> http://tranquillity.ath.cx/clang/2010-04-02-1/
[09:35:10] <Dark_Shikari> http://tranquillity.ath.cx/clang/2010-04-02-1/report-86RU2v.html#EndPath
[09:35:12] <Dark_Shikari> this is definitely a real bug
[09:36:10] <Dark_Shikari> merbanan: how easy would it be to run one of those on latest x264?  I assume the reason we're getting a ton of new ffmpeg warnings is that clang has gotten better, so I figure it can't hurt to do another x264 run
[09:37:10] <Dark_Shikari> definitely a lot of real bugs in that list, from a quick random sampling.
[09:37:46] <Dark_Shikari> I can't understand why it bitches about float loop counters though...
[09:38:24] <merbanan> most of them don't trigger, but I agree that the code is wrong and crafted bitstreams would break it sometimes
[09:38:36] <merbanan> just a security thing
[09:38:58] <Dark_Shikari> why is it a security thing though?
[09:39:03] <Dark_Shikari> what's more insecure about floating point loop variables
[09:39:12] <Dark_Shikari> http://tranquillity.ath.cx/clang/2010-04-02-1/report-zd6QfP.html#EndPath <---holy fuck how did something this stupid get in the code
[09:39:31] <Dark_Shikari> there's even a comment saying it's supposed to work, but it doesn't.
[09:39:32] <merbanan> the floats are not exact reason
[09:40:37] <Dark_Shikari> http://tranquillity.ath.cx/clang/2010-04-02-1/report-ipOdkV.html#EndPath <---ok this is a clang bug
[09:40:40] <merbanan> I think all compilers are ok with it
[09:41:22] <merbanan> but it is an API violation
[09:41:57] <merbanan> ops :)
[09:42:02] <merbanan> 2 of me
[09:42:07] <merbanan> that cant be good
[09:42:15] <Dark_Shikari> there are WAY too many bugs in there.  seriously, post that to the ML or something
[09:43:13] <merbanan> ./configure --disable-optimizations --cc=ccc-analyzer ; scan-build -o /var/www/clang -analyze-headers make
[09:43:32] <CIA-1> ffmpeg: diego * r22788 /trunk/Changelog: minor spelling fixes
[09:43:45] <merbanan> that is what is basically needed for a ffmpeg compile run
[09:44:11] <merbanan> you need a working llvm install also
[09:44:20] <Dark_Shikari> as I said, post that to the ML
[09:44:22] <Dark_Shikari> too many bugs.
[09:45:01] <merbanan> will do
[09:46:32] <merbanan> some issues are not really bugs but it wouldn't hurt to fix those anyway
[09:46:36] <CIA-1> ffmpeg: diego * r22789 /trunk/doc/general.texi:
[09:46:36] <CIA-1> ffmpeg: Group Unix-like platforms together in the platform documentation.
[09:46:36] <CIA-1> ffmpeg: patch by Michael Kostylev, michael.kostylev gmail com
[09:54:31] <DonDiego> oh, who is going to fix those clang error reports?
[09:54:45] <Dark_Shikari> I just glanced through them
[09:54:51] <DonDiego> and?
[09:54:54] <Dark_Shikari> a quick sample of a half dozen suggests a large percentage are actual bugs
[09:55:04] <Dark_Shikari> (ignoring the "dead X" ones which are 99% useless always)
[09:55:15] <DonDiego> maintainers to the front!
[09:55:19] <Dark_Shikari> Pretty much
[09:55:28] <Dark_Shikari> a lot of them are just outright _stupid_, some are very good catches
[09:55:38] <Dark_Shikari> a "stupid" example was a function that explicitly says that it works fine on null input
[09:55:44] <Dark_Shikari> and then proceeds to call a memcpy on a null pointer
[09:55:50] <Dark_Shikari> despite the comment saying otherwise
[09:57:13] <merbanan> Dark_Shikari: I think the comment was about the format of the string
[09:57:29] <Dark_Shikari> er...
[09:57:31] <Dark_Shikari>  /* NOTE: str == NULL is accepted for an empty string */
[09:57:34] <Dark_Shikari> seems straightforward to me
[09:57:57] <Dark_Shikari> ok, so yes it will never crash on a real system
[09:57:59] <Dark_Shikari> but it's still a bug
[09:58:03] <merbanan> agree
[09:58:22] <merbanan> an if (str) should fix that
[09:58:27] <Dark_Shikari> yeah
[09:58:52] <merbanan> might even compile to the same code
[10:11:40] <Dark_Shikari> merbanan: with your command I get a lot of stuff like
[10:11:40] <Dark_Shikari> Can't exec "clang-cc": No such file or directory at /usr/bin/ccc-analyzer line 216.
[10:12:12] <merbanan> /home/banan/llvm/tools/clang/utils:/home/banan/llvm/Release/bin
[10:12:29] <Dark_Shikari> ?
[10:12:31] <merbanan> I have that in my path
[10:12:55] <merbanan> it is needed
[10:14:26] <Kovensky> Dark_Shikari: using clang+llvm svn?
[10:14:49] <Dark_Shikari> emerge
[10:15:08] <merbanan> use svn
[10:15:13] <Dark_Shikari> :effort:
[10:15:57] <Kovensky> 2.6 is too old
[10:16:07] <Kovensky> and it only produces the 'clang' binary
[10:16:21] <Kovensky> it also miscompiles ffmpeg (though svn also did last time I checked)
[10:16:30] <Dark_Shikari> 2.6 what
[10:16:49] <Dark_Shikari> oh, llvm
[10:17:17] <Kovensky> 2.7 will be released on my bday, but for now you need svn =p
[10:17:38] <Dark_Shikari> birthday?
[10:17:40] <Dark_Shikari> when's that
[10:18:30] <Kovensky> the 12th
[10:18:39] <_av500_> on the day that 2.7 comes out
[11:31:09] <pengvado> are we somehow marking the issues that have been determined to be clang's mistakes?
[12:01:14] <Compn> pengvado : you can enter each one into roundup if its valid of course ...
[12:01:27] <Compn> err
[12:01:29] <Compn> clang mistakes
[12:01:32] <Compn> no idea :)
[12:01:37] * Compn cant read
[12:04:18] <lu_zero> good morning
[12:05:43] <lu_zero> j-b_afk: yesterday I had a chat with christian racca and among the rest you went in the discussion, do you still need a server? He would be glad to help in case
[12:17:22] <CIA-1> ffmpeg: mstorsjo * r22790 /trunk/libavformat/rtsp.c: Remove a redundant assignment, found by clang
[14:15:53] <CIA-1> ffmpeg: stefano * r22791 /trunk/ (17 files in 3 dirs):
[14:15:53] <CIA-1> ffmpeg: Replace all remaining occurrences of AVERROR_NOMEM with
[14:15:53] <CIA-1> ffmpeg: AVERROR(ENOMEM).
[14:15:53] <CIA-1> ffmpeg: AVERROR_NOMEM is deprecated and will be dropped at the next libavutil
[14:15:53] <CIA-1> ffmpeg: major bump.
[15:05:06] <CIA-1> ffmpeg: vitor * r22792 /trunk/libavcodec/ (binkaudio.c fft-test.c): Make code using 1d-DCT consistent with the API change
[15:06:15] <CIA-1> ffmpeg: vitor * r22793 /trunk/libavcodec/fft-test.c: Add help for (I)RDFT test in fft-test
[15:24:09] <hyc> clang stuff sounds familiar http://www.symas.com/blog/?p=53
[15:24:32] <hyc> sometimes you don't just need smarter tools, you need smarter programmers...
[15:30:37] <CIA-1> ffmpeg: koorogi * r22794 /trunk/libavfilter/avfilter.c: Remove variable whose value was never read. Found by clang.
[15:31:40] <Kovensky> koorogi?
[15:34:40] <jai> bobby