[petsc-dev] broken builds

[Satish Balay]

On Tue, 30 Aug 2011, Jed Brown wrote:

> On Tue, Aug 30, 2011 at 14:24, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
> > The Satish way (which is always the correct way) is to do rm -rf
> > ${PETSC_ARCH} and then run ./configure again.
> 
> 
> Yes, but the run-time of this operation is 2^{1 + millions of dollars paid
> for machine} minutes so it clearly won't work at the exascale. You're going
> to be left in the past if you keep thinking like that.
> 
> 
> More seriously, rm -rf ${PETSC_ARCH} is horrible because you lose
> reconfigure-$PETSC_ARCH.py if you don't relocate it first.

Yes - this is one drawback. Also there is another issue that 'rm -rf'
adresses that --download-package=reinstall can't do easily [without a
proper traker and uninstaller]

All --download-package=reinstall can do is install the new one over
the old one. This can can still break if the new one doesn't
completely overwrite the old one.

One example: --download-superlu can change from installing
-lsuperlu4.1 to -lsuperlu4.2. Perhaps there are other suttle breakage
senarios aswell..

Satish