[MPlayer-dev-eng] [PATCH] Drop support for gcc-2.95

[Uoti Urpala]

On Mon, 2006-07-10 at 10:01 +0200, Reimar Doeffinger wrote:
> > I'd require C89 variable declaration since the code is simpler to follow
> > that way.
> 
> Not to mention that it is more like what the compiler produces anyway.
> And to support the "code is simpler to follow" argument, you know where
> to look what type a variable has instead of having to search the whole
> code block. Of course, the real "bug" is that there are such large code
> blocks, but that won't change anytime soon.
> So overall at least in this case supporting gcc 2.95 actually seems like
> a good idea.

Placing variable declarations elsewhere than at the start of the block
does in many cases produce more readable code. "You know where to look"
is a disadvantage if the alternative is having the definition as part of
the code you're looking at so you don't need to go looking elsewhere at
all.

Anyway I think such stylistic discussion is mostly beside the point.
Having style-based guidelines for that would be silly when widely
varying coding styles are otherwise allowed. The question is whether
people are supposed to keep avoiding constructs that gcc 2.95 would fail
to compile, whether they should check for such constructs before
committing or when reviewing patches. That does not happen "for free"
even if the workarounds are not especially difficult in most cases. Even
with 2.95 supposedly still supported several people have already
committed code which breaks it, and there have probably been a lot more
cases where someone has noticed it and has had to alter written code
before committing. You've yourself pointed out use of such code in
reviewed patches recently.

The people expressing support for 2.95 so far seem to basically argue
that it should be supported because support doesn't cost anything. While
the cost of support is not huge it is not insignificant either. I think
actively maintaining support for obsolete software would require better
justification.