Re: cvs commit: ports/www/akregator Makefile distinfo

From: Kris Kennaway <kris_at_obsecurity.org>
Date: Thu, 2 Dec 2004 17:22:04 -0800
On Fri, Dec 03, 2004 at 01:57:26AM +0100, Michael Nottebrock wrote:
> Kris Kennaway wrote:
> 
> 
> >touch -t should work as expected, there must have been something else
> >going on here.
> 
> Can't think of anything else.
> 
> I've encountered problems with modifying timestamps in workdirs and KDE's 
> buildsystem before (the kdevelop port was notorious for shipping broken 
> distfiles which needed surgery on the file timestamps for a while and it 
> would work for me and everyone I could get to test, but try to regenerate 
> stuff on random other people's machines and/or the package cluster).

I do see this kind of thing quite commonly on the package machines
(dctc is another egregious offender that is often broken).  It could
be some kind of a race condition, e.g. ntpd steps the time backwards
during the extraction; the files are mis-ordered in the distfile which
causes the "input" file to have a later timestamp than the "output"
file (thus make will try to rebuild it); or something more subtle [1]

Kris

[1] Another common bug I see (only on 4.x now, thanks to bsdtar) is
that gtar thinks a file "changed as we read it" during package
creation and exits with an error, when nothing should have touched
that file since installation finished, prior to the start of the
package creation phase.  I can only guess this is either a bug in
gtar, or something like softupdates updating the timestamp on the
written data after the file write returned, when the gtar process may
already be running.

Received on Fri Dec 03 2004 - 01:16:32 UTC