Re: GCC 4.0 [Re: FreeBSD 6 is coming too fast]

On Mon, Apr 25, 2005 at 08:05:14AM -0600, Scott Long wrote:
> Miguel Mendez wrote:
> >On Mon, 25 Apr 2005 07:42:54 -0600
> >Scott Long <scottl_at_samsco.org> wrote:
> >
> >
> >>>According to gcc-4.0 release notes, compilation speed for C++ was
> >>>dramatically increased, up to 25% IIRC.  I think 4.0 is good
> >>>candidate for merging into HEAD.
> >
> >
> >>Is this work that you plan on doing for us?  
> >
> >
> >Definitely not for 6.0, and I usually avoid .0 releases on critical
> >software, but nonetheless it would interesting setting a tinderbox,
> >launch a buildworld process with gcc40 and see where/if it breaks. I
> >have a spare k6-2 box I could setup for that task.
> >
> >
> >>What about the deprecated language constructs in 4.0?

gcc40 revelas some interesting "bugs" in the code...

look at hysteria.sk/~neologism/ddb.patch
which is 10 minutes try to build kernel with gcc40, look at the sigedness bug
(at the bottom of the patch)
which is not (I dont know why) revealed by current gcc we use


I dont claim the patch is absolutely ok or solves anything but its nice to see

> >According to http://gcc.gnu.org/gcc-4.0/changes.html, some of the
> >deprecated constructs are not even valid C, so I see this as an
> >opportunity to fix buggy code. 
> >
> >
> >>What about the lack of exposure that it's
> >>had outside of the FSF and Apple development circles?

I kinda like the way dfly treats this. they have/had two gccs in tree which
allowed them for more smoother transition to the newer one..

> >Exactly the reason why testing will be beneficial. The more tested the
> >product on FreeBSD the more robust it will be when it's time to get it
> >into the tree.
> >
> >Cheers,
> 
> Well, I'd caution against jumping into GCC 4.0 just because of the
> claims of 25% speed improvement.  That's about the single worst reason
> to do it.  But if you're interested in moving the technology forward,
> I'd happily encourage you.  The changed language constructs are the big
> problem (extern struct foo bar[]; is no longer valid) since you not only
> need to sweep the FreeBSD tree for them, you also need to sweep the
> ports tree.  I have mixed feeling on the value of GCC making sloppy
> language extentions available for years and then suddenly revoking them,
> and I know others that have been affected by 4.0 aren't terribly happy
> either.

we will have to import gcc4.x one day just because old gccs tends to be
abandoned and not supported anymore. and I personally think that the new gcc is
improvement... sure, some work is needed but its not that big deal as it might
seem..

just my 2 cents

roman