Skip to content

Conversation

@bwbarrett
Copy link
Member

Switch the ordering of both the -L and rpath/runpath arguments
added by the wrapper compiler so that the install's libdir
is always the first in the list of paths to search. Previously,
the component-related paths were added first, which could cause
applications to find an errant (to us, anyway) libmpi.

Signed-off-by: Brian Barrett bbarrett@amazon.com
(cherry picked from commit bed9f20)
(cherry picked from commit 99c795d)

Switch the ordering of both the -L and rpath/runpath arguments
added by the wrapper compiler so that the install's libdir
is always the first in the list of paths to search.  Previously,
the component-related paths were added first, which could cause
applications to find an errant (to us, anyway) libmpi.

Signed-off-by: Brian Barrett <bbarrett@amazon.com>
(cherry picked from commit bed9f20)
(cherry picked from commit 99c795d)
@bwbarrett bwbarrett added this to the v4.0.6 milestone Jun 2, 2021
@bwbarrett bwbarrett requested a review from jsquyres June 2, 2021 21:31
@bwbarrett
Copy link
Member Author

The cherry pick is from the v4.1 branch, since pulling directly from master would create conflicts I already fixed in the 4.1 backport.

Refs #8990

@bwbarrett bwbarrett changed the title Ensure that libdir is always first in search path v4.0: Ensure that libdir is always first in search path Jun 2, 2021
@hppritcha
Copy link
Member

i'm assuminng the github cherry pick checker is confused about having two cherry-picked from strings so ignoring this failure.

@bwbarrett
Copy link
Member Author

i'm assuminng the github cherry pick checker is confused about having two cherry-picked from strings so ignoring this failure.

No, I think it's actually doing the right thing. The second cherry pick line is from the v4.1.x branch, but the commit hasn't been pushed on that branch yet. I'll sort it out once #9028 is pushed.

@hppritcha hppritcha force-pushed the bugfix/v4.0.x-wrapper-dash-L-ordering branch from 7a0b1f2 to f027bee Compare June 3, 2021 14:21
@hppritcha
Copy link
Member

we are wanting to get a release tagged today so did some editing of the commit to fool the cherry-pick checker.

Copy link
Member

@jsquyres jsquyres left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Howard -- what is this line in the commit message:

(also from - fool cherrypicker 99c795d)

Please do not do that.

@bwbarrett bwbarrett force-pushed the bugfix/v4.0.x-wrapper-dash-L-ordering branch from f027bee to 7a0b1f2 Compare June 3, 2021 14:32
@jsquyres
Copy link
Member

jsquyres commented Jun 3, 2021

Just to be clear: the git commit checker is doing exactly what it is supposed to be doing. It is not allowing you to say that this is a cherry pick of abc123 until commit abc123 actually exists in the Open MPI repository (i.e., until the PR where it came from is merged).

@bwbarrett
Copy link
Member Author

bwbarrett commented Jun 3, 2021

I committed the v4.1 change and repushed to the original commit. Not unsurprisingly, the commit checker passes now. No need for silly hacks; the system worked as expected.

@jsquyres can you update your review?

@jsquyres
Copy link
Member

jsquyres commented Jun 3, 2021

we are wanting to get a release tagged today so did some editing of the commit to fool the cherry-pick checker.

If you guys need a 4.1 PR merged for situations like this, just ask. @bwbarrett merged #9028, which made the git cherry pick checker happy.

@hppritcha
Copy link
Member

we want to get this in and tag the branch to generate a 4.0.6rc3 tarball this evening.

@gpaulsen gpaulsen merged commit 7422d41 into open-mpi:v4.0.x Jun 3, 2021
@bwbarrett bwbarrett deleted the bugfix/v4.0.x-wrapper-dash-L-ordering branch November 30, 2021 18:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants