Re: freeze exception for gcc-4.5 (i386, amd64 only)

[Pierre Habouzit <madcoder@madism.org>]

On Mon, Aug 23, 2010 at 01:21:04PM +0200, Pierre Habouzit wrote:
> On Mon, Aug 23, 2010 at 04:05:32AM +0200, Matthias Klose wrote:
> > On 21.08.2010 14:56, Julien Cristau wrote:
> > >On Fri, Aug 20, 2010 at 19:33:12 +0200, Arthur Loiret wrote:
> > >
> > >>Now, to be clear, what nice things would gcc-4.5 bring to our users?
> > >>There is a complete list here [0], but those ones are, in my opinion,
> > >>very nice:
> > >>  - The new link time optimiser.
> > >>  - Improved C++0x support.
> > >>  - Plugins support.
> > >>
> > >My understanding is that lto in 4.5 is not quite there yet.  Not that
> > >I've tried it or anything.
> > 
> > I don't share your understanding. I tried it for some builds.
> 
> I've tried it on the work repository, and lto ICEs the compiler. Plus I
> get 2 other independant ICEs that I've not had time to reduce (hence the
> lack of bug report yet).

> Though this happens with the gcc-4.5 in unstable, I've not tried with
> the one from experimental yet.

Okay, it's not strictly speaking an ICE but I get, even with the one from
experimental:

qdb/qoutput-c.c:424:1: sorry, unimplemented: gimple bytecode streams do not support the optimization attribute

And frankly there isn't anything that surprising about the code in
question so, I'd say that LTO isn't ready yet.


Though with the gcc-4.5 from experimental, the other ICEs I have are
gone, so I'd say that gcc-4.5 from experimental is better than the one
currently in unstable.

It's just that LTO isn't that a compelling reason, it's not 100%
production ready. The plugin infrastructure is though. But you're citing
dragonegg, and last time I checked, you had to patch gcc to export one
more symbol. If you haven't applied that patch (if it's still required)
your argument is moot. But I agree the plugin infrastructure *is* a
compelling reason.


Just my 2¢ as a /user/.

-- 
·O·  Pierre Habouzit
··O                                                madcoder@debian.org
OOO                                                http://www.madism.org