> On Wed, Jun 6, 2012 at 2:49 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> > A) make buildsystem a subrepo (svn external for those familiar with svn)
> 
>   There have already been many discussions on petsc-dev about how hg subrepos 
> suck and don't satisfy all our needs.
> 
> I think "our needs" are mostly wrong in this case. ;-)
> 
> Specifically, I think it is a long-running bug that the petsc-dev repository 
> does not record when a BuildSystem change is needed. The implicitly 
> synchronized or fragile "compare dates on the commit" model is bad.

If the only issue is being able to get the BuildSystem revision associated with 
a pets-dev revision, 
How about adding a hook to BuildSystem that at each commit, writes the number 
of the revision into a file in the pets-dev repository and does a commit in the 
pets-dev repository? Then all that would be done is

cd $PETSC_DIR
hg up -r <petsc revision>
cd config/BuildSystem
hg up -r `cat $PETSC_DIR/BuildSystemVersion.txt`
Which I am sure can be done automatically with a hook.

It's not high tech, but it would work, right? Plus this could be done a 
posteriori for all past revisions of BuildSystem.

Blaise


> 
> Anyway, I thought the real concerns were that old versions of Hg didn't 
> manage subrepos consistently, not that the general model was wrong.

-- 
Department of Mathematics and Center for Computation & Technology
Louisiana State University, Baton Rouge, LA 70803, USA
Tel. +1 (225) 578 1612, Fax  +1 (225) 578 4276 http://www.math.lsu.edu/~bourdin







-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
<http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120606/0680000f/attachment.html>

Reply via email to