Subject: Re: another parallel-make race
To: None <email@example.com>
From: Christos Zoulas <firstname.lastname@example.org>
Date: 09/13/2005 15:50:42
In article <20050913142552.GA9264@spathi.chuq.com>,
Chuck Silvers <email@example.com> wrote:
>On Mon, Sep 12, 2005 at 09:38:51PM +0100, segv wrote:
>> On Mon, 12 Sep 2005 21:32:52 +0100
>> Patrick Welche <firstname.lastname@example.org> wrote:
>> > On Mon, Sep 12, 2005 at 12:48:13PM +0100, segv wrote:
>> > > On Mon, 12 Sep 2005 11:04:39 +0100
>> > > Patrick Welche <email@example.com> wrote:
>> > >
>> > >
>> > > > From my point of view the problem "went away"... I didn't actually track
>> > > > anything down..
>> > >
>> > > I don't think it went away, I have just updated src and xsrc trees for
>> > > netbsd-current. Running build.sh with -j 4 results in the
>> > > Running with -j 1 "fixes" the problem.
>> > That's what I used to get - I just tried again with -j6 and today's cvs,
>> > and didn't hit the race condition, though I did reproducibly before :-/
>> > Cheers,
>> > Patrick
>> Well I ran 'cvs update' today as well. I was building on a dual Pentium 3, I
>> guess this triggers the errors. Try building it on a SMP machine, maybe you'll
>> be able to reproduce the errors.
>nevermind, I see that luke fixed the race I was complaining about
>the same day as his previous post.
>the race that you're still seeing is different. we end up with multiple
>instances of pax operating on same destination at the same time,
>and pax is dumb about trying to create a directory where a directory
>already exists. in this case, pax removes the directory and then creates
>a new directory in the same place, which can cause problems if another pax
>is trying to create something in that directory while it's gone.
>please try the attached patch (which both fixes the above problem and
>updates pax's copy of mkpath() with the race-avoiding version from mkdir.c).
I committed it with minor changes. Many thanks.