On Tue, 2 Feb 2016, Jed Brown wrote: > Satish Balay <[email protected]> writes: > > > On Tue, 2 Feb 2016, Jed Brown wrote: > > > >> Satish Balay <[email protected]> writes: > >> > >> > Well my recommendation was for this use case of "same install on a > >> > different machine" > >> > >> If you're installing to the same location on equivalently-configured > >> machines, > > > > I presumed the copy is to a different machine - perhaps with different file > > system/organization. > > Like different versions of compilers or different library versions? > Then the failure mode I mentioned is possible.
This has nothing to do with sources in PETSC_ARCH/externalpackages. They are equivalent to tarballs. [stuff in PETSC_ARCH/include can conflict] > > >> And it breaks _constantly_. It must be like 30% of petsc-maint traffic > >> in the past year; advice being "delete PETSC_ARCH and reconfigure". > > > > All of these breakages are usually due to 2 different cases. > > - [as mentioned before] - due to 'git pull' changing dependencies > > - [without git pull or equivalent] conflict between switching from say > > --download-mpich to --download-openmpi - in install location > > [PETSC_ARCH/include] and not sources in PETSC_ARCH/externalpackages > > > > I agree starting from tarballs is the recommended way - but this copy > > [of petsc+externalpackages together] will also work. > > I said it was unreliable, not that it would necessarily fail. I don't > think we can claim this is supported behavior without a way to ensure > clean build directories. Each time configure is invoked with --download-package - it does a 'make clean' or equivalent in each of these packages/build dirs - and then rebuilds. Satish
