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

Re: [petsc-dev] petsc-dev on bitbucket
by Satish Balay 09 Feb '12

09 Feb '12

On Thu, 9 Feb 2012, Sean Farley wrote: > > > > Hg named branches are kind of screwy and bookmarks (which are less screwy) > > are still an "extension" > > > > This is true. Named branches are directly inherited from older versioning > systems (cvs, svn, etc.). After 2.0 Matt Mackall was convinced that > bookmarks so be completely analgous to git's branching: > > $ hg branch foo > marked working directory as branch foo > (branches are permanent and global, did you want a bookmark?) > > Mercurial 2.1 includes fixes that move and update bookmarks automatically > (and also allow pulling divergent bookmarks: foo@1, foo@2) > > so it makes sense to have separate clones to use for release management. > > But these releases get merged back, so just tagging them would place the > > tarball in the right place. > > > > Righto. > > > > But PETSc release tarballs include generated documentation, so using > > bitbucket's auto-generated tag-tarballs is not enough for releases. > > > > Ah, that's right. I think this could be fixed by using the tag info in the > scripts that generate the documentation. Sure alternatives are possible. I guess the primary question is: is are you suggesting this alternate workflow - just becuase its possible - or because its better? This conversation started off with solving a different problem - and then turned into changing workflow. We have a decent workflow currently - and the whole change of workflow [eventhough its possible] -I don't think buying us much. And I suspect it has additional complexity which some of us don't want to deal with. If we have to create multiple accounts to better organize repos - so be it.. Satish

1 0

Re: [petsc-dev] MatMatMult gives different results
by Satish Balay 09 Feb '12

09 Feb '12

On Thu, 9 Feb 2012, Jed Brown wrote: > On Thu, Feb 9, 2012 at 20:29, Satish Balay <balay(a)mcs.anl.gov> wrote: > > > Is it possible to add such a trigger to bitbucket? > > > > Yes > > http://confluence.atlassian.com/display/BITBUCKET/Setting+Up+the+Bitbucket+… unless I'm mistaken - it looks like https - not ssh. > > > The simple convenience of having https:// and > > > ssh:// trivially push/pull the same repository is enough to make me > > prefer > > > interacting with bitbucket (screwy networks filter ssh). > > > > Not sure what problems you had. The only known issue with ssh is - we > > have fail2ban running on petsc.cs.iit - and when there are ssh > > connections from misconfigured ip adresses [where dns lookup does not > > work - it thinks its a breakin attempt - and tries to ban that > > access]. > > > > No, the problem is goofy wireless networks. This bit Matt when he was at > Brown a few weeks ago one of matts issues was what I described [don't know if it was from brown or somewhere else] > and me in Istanbul last week. AFAIK, we have had to > decide for each repo whether to use ssh or https for pushing, so I can't hg > push https://petsc.cs.iit.edu/petsc/petsc-dev. Bitbucket supports both > simultaneously, and the user manages their credentials so you don't have to. yes - I'd rather not do https on petsc.cs.iit.edu [I could - but as you say - its combersome compared to bitbucket]. I guess the alternative - is to push to your personal bitbucket clone - and ask someone else to pull/sync to the master. [Isn't this one of the things you and Sean were advocating as advantages of bitbucket? If so - I think it should be equally useful for syncing with petsc-dev - when hosted at petsc.cs.iit.edu] Satish

1 0

Re: [petsc-dev] petsc-dev on bitbucket
by Sean Farley 09 Feb '12

09 Feb '12

> > IDENTIFICATION DIVISION. > PROGRAM-ID. YOUR-FATHERS-COBOL. > PROCEDURE DIVISION. > DISPLAY 'I do'. > STOP RUN. > And now my eyes are bleeding. Thanks.

1 0

Re: [petsc-dev] petsc-dev on bitbucket
by Sean Farley 09 Feb '12

09 Feb '12

> > Hg named branches are kind of screwy and bookmarks (which are less screwy) > are still an "extension" > This is true. Named branches are directly inherited from older versioning systems (cvs, svn, etc.). After 2.0 Matt Mackall was convinced that bookmarks so be completely analgous to git's branching: $ hg branch foo marked working directory as branch foo (branches are permanent and global, did you want a bookmark?) Mercurial 2.1 includes fixes that move and update bookmarks automatically (and also allow pulling divergent bookmarks: foo@1, foo@2) so it makes sense to have separate clones to use for release management. > But these releases get merged back, so just tagging them would place the > tarball in the right place. > Righto. > But PETSc release tarballs include generated documentation, so using > bitbucket's auto-generated tag-tarballs is not enough for releases. > Ah, that's right. I think this could be fixed by using the tag info in the scripts that generate the documentation.

1 0

Re: [petsc-dev] petsc-dev on bitbucket
by Jed Brown 09 Feb '12

09 Feb '12

On Thu, Feb 9, 2012 at 20:37, Sean Farley <sean(a)mcs.anl.gov> wrote: > Yes, but they aren't guaranteed. Lowercase is the least-common denominator > (plus who programs using capital letters?). IDENTIFICATION DIVISION. PROGRAM-ID. YOUR-FATHERS-COBOL. PROCEDURE DIVISION. DISPLAY 'I do'. STOP RUN.

1 0

Re: [petsc-dev] MatMatMult gives different results
by Jed Brown 09 Feb '12

09 Feb '12

On Thu, Feb 9, 2012 at 20:29, Satish Balay <balay(a)mcs.anl.gov> wrote: > Is it possible to add such a trigger to bitbucket? > Yes http://confluence.atlassian.com/display/BITBUCKET/Setting+Up+the+Bitbucket+… > > > The simple convenience of having https:// and > > ssh:// trivially push/pull the same repository is enough to make me > prefer > > interacting with bitbucket (screwy networks filter ssh). > > Not sure what problems you had. The only known issue with ssh is - we > have fail2ban running on petsc.cs.iit - and when there are ssh > connections from misconfigured ip adresses [where dns lookup does not > work - it thinks its a breakin attempt - and tries to ban that > access]. > No, the problem is goofy wireless networks. This bit Matt when he was at Brown a few weeks ago and me in Istanbul last week. AFAIK, we have had to decide for each repo whether to use ssh or https for pushing, so I can't hg push https://petsc.cs.iit.edu/petsc/petsc-dev. Bitbucket supports both simultaneously, and the user manages their credentials so you don't have to.

1 0

Re: [petsc-dev] petsc-dev on bitbucket
by Sean Farley 09 Feb '12

09 Feb '12

> > hmm - the url 'http://petsc.cs.iit.edu/petsc/buildsystem' doesn't work - > so > atleast apache uses case sensitive urls. Yes, but they aren't guaranteed. Lowercase is the least-common denominator (plus who programs using capital letters?).

1 0

Re: [petsc-dev] petsc-dev on bitbucket
by Jed Brown 09 Feb '12

09 Feb '12

On Thu, Feb 9, 2012 at 20:26, Satish Balay <balay(a)mcs.anl.gov> wrote: > tags are no good. we implrement branches in different clones. > > You could argue that we should throw away branches in clones have all > clones in a single branch - and change our workflow. > > But I think this will be too confusing to most of us [yeah you could > change your bash prompt to always indicate wich branch you are on - > wich is equivalent to 'cd different clone' - but not all of us are > that sophisticated] > Hg named branches are kind of screwy and bookmarks (which are less screwy) are still an "extension", so it makes sense to have separate clones to use for release management. But these releases get merged back, so just tagging them would place the tarball in the right place. But PETSc release tarballs include generated documentation, so using bitbucket's auto-generated tag-tarballs is not enough for releases.

1 0

Re: [petsc-dev] petsc-dev on bitbucket
by Satish Balay 09 Feb '12

09 Feb '12

On Thu, 9 Feb 2012, Sean Farley wrote: > > > > Isn't Matt pissed off with this? 'BuildSystem' -> 'buildsystem' > >> > >> [I thought the naming scheme with caps was important - and looks like > >> bitbucket doesn't support that] > > > > > > Who architects an entire configuration that is dependent on the file > > system? Both mac and windows, by default, are case-insensitive. Making > > something this fragile is just asking for pain later. Just search the > > mercurial and git mailing lists for unicode issues and see what a mess it > > is to rely on capitalization. > > > > Also, URLs aren't guaranteed to be case-sensitive in the standard. hmm - the url 'http://petsc.cs.iit.edu/petsc/buildsystem' doesn't work - so atleast apache uses case sensitive urls. satish

1 0

Re: [petsc-dev] petsc-dev on bitbucket
by Sean Farley 09 Feb '12

09 Feb '12

> > tags are no good. we implrement branches in different clones. > Which is not a better model. > You could argue that we should throw away branches in clones have all > clones in a single branch - and change our workflow. > Yes, I would suggest this. But I think this will be too confusing to most of us [yeah you could > change your bash prompt to always indicate wich branch you are on - > wich is equivalent to 'cd different clone' - but not all of us are > that sophisticated] I think you are misusing the term "branch". Look at the repo for mercurial: http://selenic.com/hg/tags There are no other branches (just 'default' and 'stable'). Released versions are just signed tags. This is what the mercurial community (e.g. bitbucket's implementation) takes for the default. You don't need a fancy prompt (but why *wouldn't* you want a fancy prompt??) to achieve this workflow, just move the tag to the newest bugfix and be done with it.

1 0