The MPI guidelines currently state
"If the packager wishes to provide alternatives support, it MUST be
placed in a subpackage ..."
However, it has become clear that this statement is not clear enough
and warrants an additional specification. The guidelines were written
as to be fair to all people independent of the compiler and runtime
they are using.
However, some MPI runtimes have been misbehaving by autoloading their
environment with a file in /etc/profile.d placed in the runtime
package (BZ #647147). This has disunited the user interface and caused
many problems for package builders trying to compile a serial version
of an MPI-aware code (see for instance BZ #737043).
I am thus proposing that the MPI guidelines be clarified by the
"MUST: If the maintainer wants to provide autoloading support for the
MPI environment, it must be placed in a separately installable
(see text in bold italics at
We discussed this at the meeting today and agree that the usage in the bug report goes against the spirit of the guideline. We may explore modifying the environment-modules guidelines in the future to try to make things more clear what to do when you have a default and when you do not have a default. For now we took a vote on the proposed changes to the MPI guidelines:
(+1:4, +0:0, -1:0)
We need five to pass. I'll try to get the remaining FPC members to vote in this ticket. With one more +1, this will pass.
Sorry, thought the meeting was over and was looking at some patches :)
Passed (+1:5, +0:0, -1:0)
Guideline has been updated.
The MPI Guidelines have been clarified by adding this additional statement:
If the maintainer wishes for the environment module to load automatically by use of a scriptlet in /etc/profile.d or by some other mechanism, this MUST be done in a subpackage.
Metadata Update from @spot:
- Issue assigned to toshio
to comment on this ticket.