PR 23551: why should we coalesce inlined variables?

[Andrew Pinski]

On 7/9/07, Alexandre Oliva <aoliva@redhat.com> wrote:
> Rats, I forgot one of the most important and demanding ones:
>
> - huge code bases, that take weeks to build even in build clusters:
>   seriously, tell them to rebuild with -O0 -g such that they can debug
>   and they will laugh you out and focus on the other compiler vendors
>   they're already doing business with

And now you just made someone's code even worse with respect of
compile time.  So what is it?  I don't get why you want better
debugging info but slow down compiling someone's code even though they
did not ask for debugging info?  This is the biggest problem with your
patch.

I know of code bases which takes 2 hours (and even longer with
optimization) to build and link times on this is like an half an hour
(at -O -g).  But I don't see anyone trying to get link times down
(well I tried and tried but since I am not an expert on binutils I
can't do much).  If it takes weeks to build, then we are doing
something wrong, I am serious at that.  Usually -O0 -g builds are much
faster than -O2 -g builds anyways.  If they laugh at us, then we
should speed up the compiler instead of improve debugging info :) and
worry about better debugging info for optimized builds later.

Also the code bases I am talking about are all PS3 games.  Yes I am
hearing these complaints.  The compile/debug time is too high for most
of them.  They are comparing us to the Xbox 360's compiler.  Now you
can say this is my problem only but it is all of GCC issue too because
we (Sony) are exposing them to GCC and would like them to want more.

Thanks,
Andrew Pinski