[petsc-dev] moab nightlybuild failure

[Jed Brown]

Satish Balay <balay at mcs.anl.gov> writes:

> For one - I don't know why the GNUPackage requires URLs to be split up as such:
>>>>>>
>     self.downloadpath      = 'http://ftp.mcs.anl.gov/pub/fathom/'
>     self.downloadname      = 'moab'
>     self.downloadfilename  = 'moab'
>     self.downloadversion   = 'nightly'
>     self.downloadext       = 'tar.gz'

This seems overly verbose to me, though note that tarballs can unpack
into a different directory.

> But ignoring this we currently have stuff like the following - where
> the urls are attempted in the sequence specified. If the first one
> fails - try the next one:
>
>>>>>>
>     self.download   = ['https://elemental.googlecode.com/files/elemental-0.79-p1.tgz',
>                        'http://ftp.mcs.anl.gov/pub/petsc/externalpackages/elemental-0.79-p1.tgz']
> <<<<<<
>
> with [potential support] for multiple protocols. - as retrieval.py lists:
> [we also support urls of type file://path]
>
>>>>
> # Fix parsing for nonstandard schemes
> urlparse.uses_netloc.extend(['bk', 'ssh', 'svn'])
> <<<<
>
> And currently this default can be overwriten by user with:
> --download-package=url
>
> In this context we should not have a separate stuff for git
> [self.giturls/self.gitcommit]. It should be rolled into the above
> common url notation.

Maybe, though the download and 

> [note: the current git code ignores user specified
> --download-package=url - and just goes ahead and does the git stuff
> specified in package.py]

Git urls don't necessarily look different from archives, but they need
to be fetched differently.  Also, it may be useful to name more than one
mirror, but it's nice to not need to repeat the tag name/commit hash.

But I guess we could make our own URL schema.  When I looked at
retrieval.py, I found the following and didn't bothering adding new code
there.

  ##############################################
  # This is the old shit
  ##############################################

It looks like we could extend the URL schema to support the gitcommit.

> Somehow we have to add in 'prereqisites' to the 'repo' type urls in
> the above scheme.
>
> And then perhaps retrieval.py can be asked to ignore repo-urls with
> some option like '--url-git=no'

Seems awfully messy.

> Also - the repo urls need better support wrt: 'getting updates &
> rebuilding package'.  Current code for this is tarball centric [with
> some bad behavior - like not detecting a changed url etc..]

Intent of gitcommit was to have something to cache so that we would know
when the file had changed (either forward or backward).
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130627/9d9f89db/attachment.sig>