[distcc] Re: PATH problem 2

Wayne Davison wayned at users.sourceforge.net
Tue Jul 8 05:53:36 GMT 2003


On Tue, Jul 08, 2003 at 01:28:16PM +1000, Martin Pool wrote:
> On  4 Jul 2003, Leif Huhn <leif at dkstat.com> wrote:
> > Ok...this is the problem.  Maybe dcc_trim_path should just remove
> > the ccache-masq directory instead of it and everything before it?
> 
> Wayne, do you think that would work?

That would be an acceptable change for dealing with a ccache masquerade
dir, but not for dealing with a distcc masquerade dir.  However,
becoming more convinced that having the _daemon_ pre-twiddle the path is
a bad idea.

Perhaps it would be better to have it scan the path and just output
warnings if it finds a ccache masquerade dir and/or a distcc masquerade
dir -- this leaves the ball in the users court to either fix the problem
or to ignore it.  We could have the code suggest setting DISTCCD_PATH,
and if that were set, it would be used without being scanned (so that
the user had a way to silence any warnings if they wanted to use ccache
in the path).

After all, the daemon code will successfully ignore one level of distcc
loopback -- i.e. if the daemon calls a distcc symlink, it will force a
local compile that will succeed as long as it doesn't find distcc a
second time.

If you still want actual trimming, I'd like to see a part of the patch
that Dimitri sent back in May applied to the path-trimming code:
making it stop if it finds a NON-symlink on the path before finding a
masquerade dir would make it work a little better in the case where the
compiler is early in the path (before the ccache masquerade dir for
some strange reason).  I don't want the part of the patch that caused it
to quit the scan at the first unknown symlink though -- that would make
it fail to get past a colorgcc masquerade dir, for instance.

..wayne..



More information about the distcc mailing list