[PETSc] Add builder #845
[PETSc] Add builder #845
Conversation
|
Just by looking to the list of platforms that succeeded, my guess is that the build script is trying to execute a program that has built with the cross-compiler |
I am trying to figure out why they are failing, but the details are stored in a log file after running Do you have any recommendations to display the log file in the terminal of the different platforms if the config run doesn't succeed? |
The issue is the use of SOWING which is mandatory apparently. |
You can There may be a way to not make the entire bash script exit on failure of an individual command. Perhaps the opposite of |
|
It looks like SOWING produces the man pages. Is there a |
It doesn't like it. |
|
What if we build with |
|
Some progress, in windows now is complaining because of this @giordano Is there a fix or shall we remove the shared libraries in windows? |
This worked for me locally 🤞 this will work on the CI |
Co-Authored-By: Mosè Giordano <[email protected]>
|
@ViralBShah neither me or @giordano can reproduce the windows error locally to check at the logs. Any ideas? |
|
@giordano and @simonbyrne ready to merge. |
|
I can't reproduce the issue locally I get
|
|
The error on MacOS is: any ideas? |
|
From what I can tell, the problem appears to be that the default linker function: |
|
It wouldn't be the first bug we find and fix in their build system... |
|
I got this in |
|
It's really weird, since the but I don't see any file with that |
|
@simonbyrne @giordano I removed adding superlu. After researching more it doesn't look like there are significant advantages without Superlu Dist and at least from searching online it is not supported out-of-the-box to integrate superluMT. We can merge this so it can be used in Sundials |
|
Does that effect the ability to use SuperLUMT in Sundials? |
No, PETSc in Sundials enables another Non-linear solver. SuperLUMT in Sundials is used in the linear solver. I want to enable this workflow: https://github.com/LLNL/sundials/blob/master/examples/ida/petsc/idaHeat2D_petsc_snes.c |
|
@ViralBShah this PR is passing so we can incorporate it with #995 |
|
Woohoo! is this ready to go? |
|
Good to go? |
|
@giordano yes. Please see the note above about SuperLU. |
|
Let's merge and see what happens? |
|
Let's go ahead, I built this with the objective of integrating it into Sundials so we will learn how did it go. |
|
Thanks @jd-lara! |
|
Fingers crossed! |
|
getting a failure on windows: A simple function to replicate this is: using PETSc_jll
function dtype(name::AbstractString)
dtype_ref = Ref{Cint}()
found_ref = Ref{Cint}()
ccall((:PetscDataTypeFromString, libpetsc), Cint,
(Cstring, Ptr{Cint}, Ptr{Cint}),
name, dtype_ref, found_ref)
return found_ref[] == 1 ? dtype_ref[] : nothing
end
dtype("Real") |
|
Does it fail only on v1.5+? |
|
Passes on 1.3, fails on 1.4+: https://github.com/JuliaParallel/PETSc.jl/runs/1024945378 |
|
Oops, I missed the 1.4 job, but yeah, I noticed it works with v1.3 |
|
(I added 1.4 to the build matrix to check) |
@ViralBShah this is the first attempt. It worked locally for Linux 64 bit. Probably will fail in windows and require some iterations to get it right.
Windows instructions
https://www.mcs.anl.gov/petsc/documentation/installation.html#windows