Ok, indeed, after digging in the code, I see that the install_step does
delete the install directory before installation. This breaks an
installation of PETSc which would use the "--download" options.
sourceinstall = True does prevent this... but it requires to install in
the source directory.
I made a pull request to modify petsc.py to give the option to perform a
download install. The pull request is here:
https://github.com/hpcugent/easybuild-easyblocks/pull/1114
We can move further discussion to the PR. Posting here just so that
people who may be interested can follow.
Maxime
On 17-02-22 18:04, Maxime Boissonneault wrote:
Some more info I discovered.
The files that PETSc downloads are actually correctly installed in the
final installation directory during the configure phase. However,
somewhere between the configure and the install phase, something (I
suspect EasyBuild) deletes them. Is it possible that EasyBuild deletes
the content of the installation directory between the configure and
the install phase ?
Maxime
On 17-02-22 14:13, Maxime Boissonneault wrote:
On 17-02-22 14:08, Kenneth Hoste wrote:
Hi Maxime,
On 22/02/2017 17:22, Maxime Boissonneault wrote:
Actually, looking through the log file, I see PETSc which says to
run make install, but easybuild runs "make" without the install
option:
3693 make[2]: Leaving directory
'/dev/shm/mboisson/PETSc/3.7.5/iompi-2016.4/petsc-3.7.5'
3694 =========================================
3695 make[1]: Leaving directory
'/dev/shm/mboisson/PETSc/3.7.5/iompi-2016.4/petsc-3.7.5'
3696 Now to install the libraries do:
3697 make
PETSC_DIR=/dev/shm/mboisson/PETSc/3.7.5/iompi-2016.4/petsc-3.7.5/
PETSC_ARCH=arch-linux2-c-opt install
3698 =========================================
3699
3700 == 2017-02-22 15:08:25,451 run.py:443 INFO cmd " make
PETSC_DIR=/dev/shm/mboisson/PETSc/3.7.5/iompi-2016.4/petsc-3.7.5/ "
exited with exitcode 0 and output:
3701 make[1]: Entering directory
'/dev/shm/mboisson/PETSc/3.7.5/iompi-2016.4/petsc-3.7.5'
It runs "make install" later, but not with the arguments that PETSc
instructs to use:
Please open an issue on this, or a PR to the PETSc easyblock to fix
this issue.
Will do... as soon as I figure out a fix. Even when running things
manually I run into problems. Not sure why yet.
Maxime
--
---------------------------------
Maxime Boissonneault
Analyste de calcul - Calcul Québec, Université Laval
Président - Comité de coordination du soutien à la recherche de Calcul Québec
Team lead - Research Support National Team, Compute Canada
Instructeur Software Carpentry
Ph. D. en physique
