This is the mail archive of the
mailing list for the GCC project.
Re: [wwwdocs] IPA and LTO updates
* Andi Kleen wrote on Tue, Nov 16, 2010 at 12:03:33PM CET:
> On Mon, Nov 15, 2010 at 11:18:42PM +0100, Ralf Wildenhues wrote:
> > * Jan Hubicka wrote on Mon, Nov 15, 2010 at 11:16:43PM CET:
> > > >
> > > > I think deadlock is only likely when you do not implement
> > > > the jobserver protocol correctly.
> > >
> > > We still eventually call make to connect to he jobserver, so there
> > > should not be problems with jobserver protocol as we implement none of
> > > it?
> > Good point.
> Ralf, you originally added the patch to not pass through the jobserver environment
> variables. What was the original reason for that?
If you pass the --jobserver-fd=... flags but the fd's are not actually
open, the sub-'make -jN' prints warnings. So, lto-wrapper should only
allow through the flags if it also passes the file descriptors (and of
course they are passed to lto-wrapper in the first place).
Another issue that is at least debatable is whether you want to pass on
flags like n (--dry-run, probably yes), k (--keep-going, not so sure),
t (--touch, not sure either). There are more.