Criteria for a warning to be in -Wall? (was: Re: a warning to implement)

[Joe Buck]

> <<I'd love a "perfect debugging as with -g only but with flow analysis
> for uninitialized warnings" option, but it's probably not easy to add.
> >>
> 
> What I really want is "perfect debugging with as much optimization as is
> possible without compromising debugging information. One of the problems
> with gcc for me is that the -O0 mode is really awful (we actually run into
> trouble with customers who insist on benchmarking unoptimized code and gcc
> is worse than any other compiler around in this mode -- yes yes we know
> about telling people that this is a good thing, that gcc really obeys
> you when you say no optimziation, but the fact is that the code is 
> unnceessarily (from a debugging point of view) awful in -O0 mode.

It seems to me that if we had an optimization level (call it -O0.5, or
just adjust -O1) in which, at the end of each line, values get flushed
back to memory, we'd pretty much have it: complicated expressions inside
one line of code could be optimized as much as you want, but the 
developer in the debugger would see just what he wrote.