[petsc-dev] --download-ptscotch

[Satish Balay]

I was thinking more in terms of TCP 45 sec timeout - but most timeout
values listed at stacexchange were smaller than that - so I settled
with 30. 10 might also be fine..

Satish

On Thu, 25 Jun 2015, Barry Smith wrote:

> 
>    Looks reasonable to me.
> 
>   30 seconds seems a bit long; if it has not made a successful connection in say 10 seconds it seems unlikely 20 more will help?
> 
> 
> 
>   Barry
> 
> > On Jun 25, 2015, at 12:56 PM, Satish Balay <balay at mcs.anl.gov> wrote:
> > 
> > Or perhaps the following [set timeout before urlretrieve - and reset after]
> > 
> > Satish
> > 
> > -----
> > $ git diff |cat
> > diff --git a/config/BuildSystem/retrieval.py b/config/BuildSystem/retrieval.py
> > index 1875a07..0b1545e 100644
> > --- a/config/BuildSystem/retrieval.py
> > +++ b/config/BuildSystem/retrieval.py
> > @@ -4,6 +4,8 @@ import os
> > import urllib
> > import urlparse
> > import config.base
> > +import socket
> > +
> > # Fix parsing for nonstandard schemes
> > urlparse.uses_netloc.extend(['bk', 'ssh', 'svn'])
> > 
> > @@ -56,8 +58,12 @@ class Retriever(logger.Logger):
> >       os.unlink(localFile)
> > 
> >     try:
> > +      sav_timeout = socket.getdefaulttimeout()
> > +      socket.setdefaulttimeout(30)
> >       urllib.urlretrieve(url, localFile)
> > +      socket.setdefaulttimeout(sav_timeout)
> >     except Exception, e:
> > +      socket.setdefaulttimeout(sav_timeout)
> >       failureMessage = '''\
> > Unable to download package %s from: %s
> > * If URL specified manually - perhaps there is a typo?
> > 
> > 
> > On Thu, 25 Jun 2015, Satish Balay wrote:
> > 
> >> Looks like the following is a way to timeout urllib.urlretrieve()
> >> 
> >> diff --git a/config/BuildSystem/retrieval.py b/config/BuildSystem/retrieval.py
> >> index 1875a07..fb5059d 100644
> >> --- a/config/BuildSystem/retrieval.py
> >> +++ b/config/BuildSystem/retrieval.py
> >> @@ -4,6 +4,9 @@ import os
> >> import urllib
> >> import urlparse
> >> import config.base
> >> +import socket
> >> +socket.setdefaulttimeout(30)
> >> +
> >> # Fix parsing for nonstandard schemes
> >> urlparse.uses_netloc.extend(['bk', 'ssh', 'svn'])
> >> 
> >> 
> >> However - it appears to be a global timeout [for all sockets created from python?] Should we use this?
> >> 
> >> Satish
> >> 
> >> 
> >> On Thu, 25 Jun 2015, Satish Balay wrote:
> >> 
> >>> We use:
> >>> 
> >>>    self.download     = ['http://gforge.inria.fr/frs/download.php/file/34099/scotch_6.0.3.tar.gz',
> >>>                         'http://ftp.mcs.anl.gov/pub/petsc/externalpackages/scotch_6.0.3.tar.gz']
> >>> 
> >>> 
> >>> So - if the first url fails - configure should use the second one.
> >>> 
> >>> However sometimes configure hangs. I see configure is hanging here -
> >>> but weget is not. So perhaps a bug in python urllib?
> >>> 
> >>> [have to check more..]
> >>> 
> >>> Satish
> >>> 
> >>> 
> >>> On Thu, 25 Jun 2015, Barry Smith wrote:
> >>> 
> >>>> 
> >>>>   Bug report should be sent to the Inria folks.
> >>>> 
> >>>>   Barry
> >>>> 
> >>>> I wonder how many French Francs they waste maintaining their own source forge instead of using github for free.
> >>>> 
> >>>> 
> >>>>> On Jun 25, 2015, at 9:55 AM, Stefano Zampini <stefano.zampini at gmail.com> wrote:
> >>>>> 
> >>>>> It seems that —download-ptscotch is broken right now, since the code repository on gforge.inria.fr is not accessible anymore as an anonymous user.
> >>>>> 
> >>>>> Stefano
> >>>> 
> >>>> 
> >>> 
> >> 
> 
>