guix-patches
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[bug#50080] Commit updating SUNDIALS to 6.1.1.


From: Paul A. Patience
Subject: [bug#50080] Commit updating SUNDIALS to 6.1.1.
Date: Tue, 26 Apr 2022 19:47:12 +0000

Hi Ludovic,

On 2022-04-15 04:50:39-04:00, Ludovic Courtès wrote:
> "Paul A. Patience" <paul@apatience.com> skribis:
>> Do you know if the update to 6.1.1 renders the issues extraneous?
>
> Thanks for bringing it up, I had overlooked this patch of yours—my
> apologies for that.  I don’t know if the new version defaults to 64-bit
> or 32-bit indices.

Ok, I figured out what happened.

--8<---------------cut here---------------start------------->8---
      ("petsc" ,petsc-openmpi)         ;support in SUNDIALS requires MPI
-       ,@(package-inputs sundials)))
+       ("petsc-openmpi" ,petsc-openmpi)      ;support in SUNDIALS requires MPI
+       ,@(alist-delete "petsc" (package-inputs sundials))))
     (arguments
      (substitute-keyword-arguments (package-arguments sundials)
        ((#:configure-flags flags '())
-        `(cons* "-DMPI_ENABLE:BOOL=ON"
-                "-DPETSC_ENABLE:BOOL=ON"
-                (string-append "-DPETSC_INCLUDE_DIR="
-                               (assoc-ref %build-inputs "petsc")
-                               "/include")
-                (string-append "-DPETSC_LIBRARY_DIR="
-                               (assoc-ref %build-inputs "petsc")
-                               "/lib")
-                ,flags))
+        `(cons* "-DENABLE_MPI:BOOL=ON" ,flags))
--8<---------------cut here---------------end--------------->8---

You accidentally disabled PETSc support in SUNDIALS by not re-adding
PETSC_ENABLE in the sundials (not sundials-openmpi) package.
If you had done so, the configure step would have failed because PETSc
is supported in SUNDIALS only with MPI (hence the comment).
Once we enable PETSc again, the index size problem appears again, even
with SUNDIALS 6.1.1.

Basically, I think all of my changes are still relevant.
I still think setting the sundials and sundials-openmpi's index sizes to
32 bits is the most logical option (see the commit message in the
“gnu: sundials: Update to 5.7.0.” patch).

I will bring my old patch series up to date and submit it again when
it's done.

Best regards,
Paul






reply via email to

[Prev in Thread] Current Thread [Next in Thread]