[petsc-dev] configure error

[Satish Balay]

I don't understand the batch code. I suspect the change from building
packages from pre-reconfigure to reconfigure happened in
package.py. However blaslapack doesn't use package.py?

And blaslapack.py appears to do checksdotreturnsdouble() test - which
it needs to do during batch test - so it would have to do the install
before batch test [and not during reconfigure?]

Sure - it shouldn't chug along...

Satish

On Thu, 29 Dec 2016, Barry Smith wrote:

> 
> > On Dec 29, 2016, at 10:03 AM, Satish Balay <balay at mcs.anl.gov> wrote:
> > 
> > I suspect blaslpack along with mpi cannot be installed in batch mode anymore.
> 
>    This is nuts!  It should be able to build blaslapack. And it certainly shouldn't just chug along without a big error message instead of pretending the option was not set.
> 
>    Presumably batchBodies is only set before the reconfigure?
> 
>    So here it skips the download before, that is ok,
> 
>   def checkDownload(self):
>     '''Check if we should download the package, returning the install directory or the empty string indicating installation'''
>     if not self.download:
>       return ''
>     if self.framework.batchBodies:
>       return
> 
>    Here it skips configure, and if this is only the first time it is ok
> 
>   def configure(self):
>     if self.download and self.argDB['download-'+self.downloadname.lower()] and not self.framework.batchBodies:
>       self.argDB['with-'+self.package] = 1
> 
>    These are the only two mentions of batch in package.py so maybe batchBodies is always set? Then we need a different way
> to know if it is before or after the reconfigure.
> 
>  Barry
> 
> 
> > 
> > [at some point all pacakges were installed before reconfigure. And I think this was
> > switched to - install packages during reconfigure. Ideally there should be proper
> > dependencies on the 'batch' part - so that only packages that require info from the
> > batch run should get installed during reconfigure stage..]
> > 
> > BTW: Why use --download-fblaslapack on cray? [and not system blas - like cray-libsci]
> > 
> > Satish
> > 
> > On Thu, 29 Dec 2016, Mark Adams wrote:
> > 
> >> --with-batch=0 seems to fix the problem.
> >> 
> >> On Thu, Dec 29, 2016 at 4:40 AM, Mark Adams <mfadams at lbl.gov> wrote:
> >> 
> >>> BTW, I cloned this (working) config file and changed to single precision
> >>> and 32 bit ints, and added with-batch=1.
> >>> 
> >>> On Wed, Dec 28, 2016 at 11:20 PM, Satish Balay <balay at mcs.anl.gov> wrote:
> >>> 
> >>>> I can reproduce this issue. [we don't notice it normally because
> >>>> default system blas is usually found - so there is no error..]
> >>>> 
> >>>> Satish
> >>>> 
> >>>> On Wed, 28 Dec 2016, Barry Smith wrote:
> >>>> 
> >>>>> 
> >>>>>  Mark,
> >>>>> 
> >>>>>   Yes this is odd. You clearly have --download-fblaslapack=1  but then
> >>>> in the file it has
> >>>>> 
> >>>>> TEST alternateConfigureLibrary from config.packages.fblaslapack(/g
> >>>> lobal/u2/m/madams/petsc/config/BuildSystem/config/package.py:761)
> >>>>> TESTING: alternateConfigureLibrary from config.packages.fblaslapack(co
> >>>> nfig/BuildSystem/config/package.py:761)
> >>>>>  Called if --with-packagename=0; does nothing by default
> >>>>> 
> >>>>>   I suggest try removing all options except the
> >>>> --download-fblaslapack=1 and the compiler names and the --with-batch and
> >>>> see if it goes through. Do not cut and past but type each argument; it
> >>>> could be some strange character being pass in you cannot see.
> >>>>> 
> >>>>>   Barry
> >>>>> 
> >>>>> 
> >>>>>> On Dec 28, 2016, at 9:52 PM, Mark Adams <mfadams at lbl.gov> wrote:
> >>>>>> 
> >>>>>> I am getting an error about blas-lapack, suggesting that I download
> >>>> it, but I seem to be asking to download it. Not sure what the problem is ...
> >>>>>> 
> >>>>>> <configure.log>
> >>>>> 
> >>>>> 
> >>>> 
> >>>> 
> >>> 
> >> 
> > 
> 
>