June 2012 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] Fwd: [petsc-maint #119133] petsc-dev configure crash
by Sean Farley 06 Jun '12

06 Jun '12

> Jed and Sean's email bitched about the current system (which is fine to bitch) but did not have specific suggestions on a new system. So how do you propose to (1) not have the "pull BuildSystem" problem and (2) have bisection work? > > Once we know the proposal we can determine the downside to the proposal and decide whether to change. But without a specific proposal (state it) you all are just flapping your lips. There are only two ways that I know of that satisfy both (1) and (2) (tell me if anyone knows of any more): A) make buildsystem a subrepo (svn external for those familiar with svn) B) merge buildsystem into petsc-dev and use graft / transplant to maintain a separate repo

1 0

Re: [petsc-dev] Fwd: [petsc-maint #119133] petsc-dev configure crash
by Gerard Gorman 06 Jun '12

06 Jun '12

Barry Smith emailed the following on 06/06/12 17:30: > On Jun 6, 2012, at 11:26 AM, Jed Brown wrote: > >> On Wed, Jun 6, 2012 at 11:24 AM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: >> With all the other unneeded bells and whistles in hg, why doesn't it have a mechanism where WE can put this post-pull business into the repository instead of telling each user to do that? >> >> Massive security hole? >> >> Sometimes I like to be able to look at code without running it. I've seen "make" run "rm -rf ..". Just because I don't trust someone doesn't mean I don't want to look at their code. It would be a very bad thing for Hg to run arbitrary code when someone clones. > Did I ever say a mechanism to "run arbitrary code"? I do not believe I did, nor did I even hint at running arbitrary code. What I want is a mechanism to run another hg command, in fact a specific hg command. Not "arbitrary code". > > Barry > > I often get tripped up by this and can understand how it can get annoying. In svn land this issue would be formally handled by externals, but the concept doesn't really carry to mercurial. One conclusion on stackoverflow was just to pull in the external repository entirely: http://stackoverflow.com/questions/217523/can-i-emulate-svnexternals-using-… Likewise, for our codes when we used svn we did use externals. But they are a major pain because you have to manage the comparability between the different repository revisions - making it awkward to roll back the code to an early release when bug tracking for example. Eventually we gave up trying to manage the insanity and just pulled in any external repositories and merge in diff's to these externals when required/desired. If pulling in the external wasn't an option, I like the second suggestion on stackoverflow where they suggest coding it into the build process. If that also isn't an option, then perhaps a warning from configure could be useful. Cheers Gerard

1 0

Re: [petsc-dev] Fwd: [petsc-maint #119133] petsc-dev configure crash
by Barry Smith 06 Jun '12

06 Jun '12

On Jun 6, 2012, at 11:41 AM, Sean Farley wrote: >> Did I ever say a mechanism to "run arbitrary code"? I do not believe I did, nor did I even hint at running arbitrary code. What I want is a mechanism to run another hg command, in fact a specific hg command. Not "arbitrary code". > > Which boils down to arbitrary code :-) Your type of proposal has > rejected repeatedly and no one has proposed a safe way to do it. If > you feel this is easy, then by all means submit a patch to mercurial. > You might as well train people to use 'hg pull -u' and have > ./configure add the hook but Satish maintains that this doesn't solve > all problems (which he is right) so it's useless (which I disagree > with). Oh well. What do you propose, Barry? I propose what we do currently. It is not perfect but at worst requires an email of "pull BuildSystem". Jed and Sean's email bitched about the current system (which is fine to bitch) but did not have specific suggestions on a new system. So how do you propose to (1) not have the "pull BuildSystem" problem and (2) have bisection work? Once we know the proposal we can determine the downside to the proposal and decide whether to change. But without a specific proposal (state it) you all are just flapping your lips. Barry

1 0

Re: [petsc-dev] Fwd: [petsc-maint #119133] petsc-dev configure crash
by Sean Farley 06 Jun '12

06 Jun '12

> Did I ever say a mechanism to "run arbitrary code"? I do not believe I did, nor did I even hint at running arbitrary code. What I want is a mechanism to run another hg command, in fact a specific hg command. Not "arbitrary code". Which boils down to arbitrary code :-) Your type of proposal has rejected repeatedly and no one has proposed a safe way to do it. If you feel this is easy, then by all means submit a patch to mercurial. You might as well train people to use 'hg pull -u' and have ./configure add the hook but Satish maintains that this doesn't solve all problems (which he is right) so it's useless (which I disagree with). Oh well. What do you propose, Barry?

1 0

Re: [petsc-dev] Fwd: [petsc-maint #119133] petsc-dev configure crash
by Barry Smith 06 Jun '12

06 Jun '12

On Jun 6, 2012, at 11:26 AM, Jed Brown wrote: > On Wed, Jun 6, 2012 at 11:24 AM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > With all the other unneeded bells and whistles in hg, why doesn't it have a mechanism where WE can put this post-pull business into the repository instead of telling each user to do that? > > Massive security hole? > > Sometimes I like to be able to look at code without running it. I've seen "make" run "rm -rf ..". Just because I don't trust someone doesn't mean I don't want to look at their code. It would be a very bad thing for Hg to run arbitrary code when someone clones. Did I ever say a mechanism to "run arbitrary code"? I do not believe I did, nor did I even hint at running arbitrary code. What I want is a mechanism to run another hg command, in fact a specific hg command. Not "arbitrary code". Barry

1 0

Re: [petsc-dev] Fwd: [petsc-maint #119133] petsc-dev configure crash
by Sean Farley 06 Jun '12

06 Jun '12

> With all the other unneeded bells and whistles in hg, why doesn't it have a mechanism where WE can put this post-pull business into the repository instead of telling each user to do that? We've been over this before. It would be a HUGE security risk. Look at the projrc extension if you really want this but that would still require each user to install the extension.

1 0

Re: [petsc-dev] Fwd: [petsc-maint #119133] petsc-dev configure crash
by Jed Brown 06 Jun '12

06 Jun '12

On Wed, Jun 6, 2012 at 11:24 AM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > With all the other unneeded bells and whistles in hg, why doesn't it have > a mechanism where WE can put this post-pull business into the repository > instead of telling each user to do that? Massive security hole? Sometimes I like to be able to look at code without running it. I've seen "make" run "rm -rf ..". Just because I don't trust someone doesn't mean I don't want to look at their code. It would be a very bad thing for Hg to run arbitrary code when someone clones.

1 0

Fwd: [petsc-maint #119133] petsc-dev configure crash
by Barry Smith 06 Jun '12

06 Jun '12

Begin forwarded message: > From: John Mousel <john.mousel(a)gmail.com> > Date: June 6, 2012 10:12:19 AM CDT > To: Jed Brown <jedbrown(a)mcs.anl.gov> > Cc: petsc-maint(a)mcs.anl.gov, petsc-dev(a)mcs.anl.gov > Subject: Re: [petsc-maint #119133] petsc-dev configure crash > Reply-To: petsc-maint(a)mcs.anl.gov, John Mousel <john.mousel(a)gmail.com> > > Thanks Jed. Somehow I forgot to put the post-pull in .hg/hgrc for that > build. With all the other unneeded bells and whistles in hg, why doesn't it have a mechanism where WE can put this post-pull business into the repository instead of telling each user to do that? Barry > > On Wed, Jun 6, 2012 at 10:03 AM, Jed Brown <jedbrown(a)mcs.anl.gov> wrote: > >> You need to pull BuildSystem. >> >> >> On Wed, Jun 6, 2012 at 10:00 AM, John Mousel <john.mousel(a)gmail.com>wrote: >> >>> I'm getting the following after pulling and reconfiguring: >>> >>> >>> =============================================================================== >>> Configuring PETSc to compile on your >>> system >>> >>> =============================================================================== >>> >>> =============================================================================== >>> >>> It appears you do not have valgrind installed on your >>> system. >>> >>> We HIGHLY recommend you install it from www.valgrind.org >>> Or install valgrind-devel or equivalent using your package >>> manager. >>> >>> Then rerun >>> ./configure >>> >>> =============================================================================== >>> >>> TESTING: configureRegression from >>> PETSc.utilities.Regression(config/PETSc/utilities/Regression.py:32) >>> >>> ******************************************************************************* >>> CONFIGURATION CRASH (Please send configure.log to >>> petsc-maint(a)mcs.anl.gov) >>> >>> ******************************************************************************* >>> >>> John >>> >>> >> >

1 0

Re: [petsc-dev] [petsc-maint #119133] petsc-dev configure crash
by John Mousel 06 Jun '12

06 Jun '12

Thanks Jed. Somehow I forgot to put the post-pull in .hg/hgrc for that build. On Wed, Jun 6, 2012 at 10:03 AM, Jed Brown <jedbrown(a)mcs.anl.gov> wrote: > You need to pull BuildSystem. > > > On Wed, Jun 6, 2012 at 10:00 AM, John Mousel <john.mousel(a)gmail.com>wrote: > >> I'm getting the following after pulling and reconfiguring: >> >> >> =============================================================================== >> Configuring PETSc to compile on your >> system >> >> =============================================================================== >> >> =============================================================================== >> >> It appears you do not have valgrind installed on your >> system. >> >> We HIGHLY recommend you install it from www.valgrind.org >> Or install valgrind-devel or equivalent using your package >> manager. >> >> Then rerun >> ./configure >> >> =============================================================================== >> >> TESTING: configureRegression from >> PETSc.utilities.Regression(config/PETSc/utilities/Regression.py:32) >> >> ******************************************************************************* >> CONFIGURATION CRASH (Please send configure.log to >> petsc-maint(a)mcs.anl.gov) >> >> ******************************************************************************* >> >> John >> >> >

1 0

Re: [petsc-dev] [petsc-maint #119133] petsc-dev configure crash
by Jed Brown 06 Jun '12

06 Jun '12

You need to pull BuildSystem. On Wed, Jun 6, 2012 at 10:00 AM, John Mousel <john.mousel(a)gmail.com> wrote: > I'm getting the following after pulling and reconfiguring: > > > =============================================================================== > Configuring PETSc to compile on your > system > > =============================================================================== > > =============================================================================== > > It appears you do not have valgrind installed on your > system. > > We HIGHLY recommend you install it from www.valgrind.org > Or install valgrind-devel or equivalent using your package > manager. > > Then rerun > ./configure > > =============================================================================== > > TESTING: configureRegression from > PETSc.utilities.Regression(config/PETSc/utilities/Regression.py:32) > > ******************************************************************************* > CONFIGURATION CRASH (Please send configure.log to petsc-maint(a)mcs.anl.gov > ) > > ******************************************************************************* > > John > >

1 0