[petsc-dev] [petsc-maint #119133] petsc-dev configure crash

[Barry Smith]

On Jun 6, 2012, at 11:13 PM, Sean Farley wrote:

> Thats not possible. If petsc-dev needed 'joe' then you would have needed to commit there (and would have either gotten an abort message about uncommitted changes or else it would be working; hence no commit necessary)

   Completely wrong. hg is magic and looks through all the source code and sees that I have referenced joe in petsc-dev and then deduces that joe must be down in BuildSystem and checks for it there and then generates a useful error message?   
 
    The problem is real but but Jed may be right that it is a minor thing that we could live with.

   Barry

> 
> On Wednesday, June 6, 2012, Barry Smith wrote:
> 
> On Jun 6, 2012, at 9:42 PM, Satish Balay wrote:
> 
> > On Wed, 6 Jun 2012, Barry Smith wrote:
> >
> >>   Satish,
> >>
> >>     You will not just "recheck this". Please setup the petsc-buildsystem-dev test repository so EVERYONE can test it and make sure there are no kinks.
> >
> > Added.
> >
> > ssh://petsc@petsc.cs.iit.edu//home/petsc/petsc-buildsystem-dev
> 
>   Thanks
> 
> barry-smiths-macbook-pro:petsc-buildsystem-dev barrysmith$ hg commit
> abort: uncommitted changes in subrepo config/BuildSystem
> (use --subrepos for recursive commit)
> 
> This is good.
> 
> I don't like the fact that running hg in the subrepo only affects that subrepo (because I generally just run hg in whatever strange place I am and assume it is going to do the right thing). I think running it in that subrepo directory should behave the same as outside it. For example
> 
> If I do
> 
> barry-smiths-macbook-pro:petsc-buildsystem-dev barrysmith$ cd config/BuildSystem/
> barry-smiths-macbook-pro:BuildSystem barrysmith$ touch joe
> barry-smiths-macbook-pro:BuildSystem barrysmith$ hg add joe
> adding joe
> barry-smiths-macbook-pro:BuildSystem barrysmith$ hg commit
> joe
> committed changeset 2830:03ab1f7aeb3a
> barry-smiths-macbook-pro:BuildSystem barrysmith$ hg push
> pushing to ssh://petsc@petsc.cs.iit.edu//home/petsc/BuildSystem-subrepo
> running ssh petsc at petsc.cs.iit.edu 'hg -R /home/petsc/BuildSystem-subrepo serve --stdio'
> searching for changes
> 1 changesets found
> remote: adding changesets
> remote: adding manifests
> remote: adding file changes
> remote: added 1 changesets with 1 changes to 1 files
> 
> or instead do this
> 
> barry-smiths-macbook-pro:BuildSystem barrysmith$ cd ..
> barry-smiths-macbook-pro:config barrysmith$ touch BuildSystem/jeff
> barry-smiths-macbook-pro:config barrysmith$ hg add BuildSystem/jeff
> adding BuildSystem/jeff
> barry-smiths-macbook-pro:config barrysmith$ hg commit
> abort: uncommitted changes in subrepo config/BuildSystem
> (use --subrepos for recursive commit)
> barry-smiths-macbook-pro:config barrysmith$ hg commit -S
> committing subrepository config/BuildSystem
> jeff
> .hgsubstate
> committed changeset 23600:007d9928edc8
> barry-smiths-macbook-pro:config barrysmith$ hg push
> pushing to ssh://petsc@petsc.cs.iit.edu//home/petsc/petsc-buildsystem-dev
> running ssh petsc at petsc.cs.iit.edu 'hg -R /home/petsc/petsc-buildsystem-dev serve --stdio'
> pushing subrepo config/BuildSystem to ssh://petsc@petsc.cs.iit.edu//home/petsc/BuildSystem-subrepo
> running ssh petsc at petsc.cs.iit.edu 'hg -R /home/petsc/BuildSystem-subrepo serve --stdio'
> searching for changes
> 1 changesets found
> remote: adding changesets
> remote: adding manifests
> remote: adding file changes
> remote: added 1 changesets with 1 changes to 1 files
> searching for changes
> 1 changesets found
> remote: adding changesets
> remote: adding manifests
> remote: adding file changes
> remote: added 1 changesets with 1 changes to 1 files
> 
> and then take a week vacation.     In the first case if someone else does hg pull in the the petsc-dev they don't get my joe file (which may be like totally needed now to use PETSc) they get all pissed off and send me email that I don't answer for a week. (Yes they could go into BuildSystem and pull but they won't think to do that).
> 
>  Am I missing understanding something. Any option in hg to get hg commands inside the subrepo behave like ones on the outside?
> 
>   Barry
> 
> 
> 
> >
> > Just to clarify - this is a play-repo - i.e all edits/clones should be
> > discarded later on.
> >
> > [I'm not sure how one would enforce this - and not have someone mix
> > commits between this repo and the regular petsc-dev So for now its not
> > available at http://]
> >
> > Satish
> >
> > ---------
>