Re: [ANNOUNCE] Sawfish 1.6.3 released!

[Allan Duncan <amd2345 fastmail com au>]

On 25/05/10 04:12, Christopher Roy Bratusek wrote:
> Am Mon, 24 May 2010 22:59:54 +1000
> schrieb Allan Duncan<amd2345 fastmail com au>:
>
> > On 24/05/10 21:41, Christopher Roy Bratusek wrote:
> > > Am Mon, 24 May 2010 20:11:38 +1000
> > > schrieb Allan Duncan<amd2345 fastmail com au>:
> > >
> > > Hi,
> > >
> > > the problem has been reported, but the cause is not so obvious at first sight, as
> > > not everyone got the issue (seems to be compile-lib-batch (…)). Apply one of the
> > > following hunks to work-around the issue:
> >
> > Do you mean Either Or of the two hunks in the patch file?
> > Applying both made the compile happy, but I'm not going to risk breaking
> > a working system by installing it.
>
> Just one of them, which one doesn't matter. You won't risk any breakage.
>
> > Which lib do you think is involved?  The system is incrementally
> > updated, so the libs used for librep on 10 Jan will almost certainly
> > changed.  I'll try recompiling librep and rep-gtk tomorrow to see if the
> > original 1.6.3 works.
>
> No, no lib. for bytecompiling, several rep-commands are available.
> compile-batch and compile-lib-batch are two of them. And it seems that
> compile-lib-batch causes this issue (somewhat randomly).

Well I tried recompiling librep and rep-gtk but the sawfish compile fell 
over on "void-value prompt-completion-fun" again.  Using the first patch 
hunk made it all beautiful.

Curious though.  Is there some interaction with multiple cores and 
parallel compiles?
I had that with something else the other day.  The makefile for lame has 
the test
    test -d $(distdir) || mkdir $(distdir)
in a loop, and there are parallel invocations of the test before the 
mkdir has happened.  It should have just been
    mkdir -p $(distdir)
anyway, no need for the test.