[MPlayer-users] ffmpeg/libavcodec

[Alex Kanavin]

On Wed, 24 Oct 2001, Arpi wrote:

> And for the people, who periodically asks what are the exact problems with
> gcc 2.96, my answer: we don't know. we just see various bugreports mostly
> gcc internal bugs, 

Not a single report of that since June 2001. But there were reports of 
gcc 3.0 segfaulting indeed.

> compiler syntax errors in source 

Not specific to 2.96 - 3.0 also had them. Be sure, I've checked all the 
reports since June 2001.

> or bad code compiled. They

Just MMX code skipping.

> all are solved using different version of gcc. I understand that gcc 2.96
> has different default optimization flags and they conflicts with our inline
> asm code, but we can't fix them, and we really don't want to fix them as they
> work with other compilers or gcc versions, and the fix may cause speed loss.

I have read the archives since June 2001. All the problems with 2.96 are 
classified into these:

1) mmx code skipping caused by pipe in comments
2) cant find a register - not specific to 2.96, 3.0 also has it.
3) the rest is NOT specific to 2.96 also - ppl tried other compilers and 
still got the bug.

So, 2.96 is not buggier than anything else. But you're not happy with it. 
One of life's unsolved mysteries...

Anyway, I feel the best solution would be to remove compiler version 
checking alltogether, but produce a warning (compile-time and run-time) 
that "if you use 2.96 we ignore you" and actually ignore 2.96 bugreports.
But it's up to you of course. 

-- 
Alexander

Homepage: http://www.sensi.org/~ak/