Bug#1042005: Info received (Bug#1042005: transition: mumps hypre2.28.0 superlu combblas)
Hi On Thu, 31 Aug 2023 at 12:03, Sebastian Ramacher wrote: > Unfortunately trilinos is a key package and so cannot be removed without > much effort from testing to complete the transition. Can you please take > a look at fixing the current issues with trilinos? As can be seen on salsa [1], packaging of trilinos 14.4.0-1 is underway, although it is taking a bit longer than expected. In the meantime, I've uploaded trilinos 13.2.0-5 to unstable, which should fix the build with GCC-13. Regards Graham [1] https://salsa.debian.org/science-team/trilinos
Bug#1042005: Info received (Bug#1042005: transition: mumps hypre2.28.0 superlu combblas)
On 2023-08-19 14:08:53 +0200, Drew Parsons wrote: > All direct uploads have been made and migrated to testing. > > trilinos is not yet rebuilt against MUMPS 5.6. I gather trilinos has other > issues not related to this transition (gcc-13 for instance). > > Apart from that I think this transition is done. Unfortunately trilinos is a key package and so cannot be removed without much effort from testing to complete the transition. Can you please take a look at fixing the current issues with trilinos? Cheers -- Sebastian Ramacher
Bug#1042005: Info received (Bug#1042005: transition: mumps hypre2.28.0 superlu combblas)
All direct uploads have been made and migrated to testing. trilinos is not yet rebuilt against MUMPS 5.6. I gather trilinos has other issues not related to this transition (gcc-13 for instance). Apart from that I think this transition is done. Drew.
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
On 2023-08-11 03:55, Drew Parsons wrote: ... Unfortunately we're getting test errors from PETSc running its superlu-dist test. ... Discussed upstream at superlu-dist issue #141 and PETSc issue #1384. https://github.com/xiaoyeli/superlu_dist/issues/141 https://gitlab.com/petsc/petsc/-/issues/1384 The superlu-dist patch is (I think) https://github.com/xiaoyeli/superlu_dist/commit/f36d65a4e3dee264a18f4bd17b3ec506173ccbc2 Unfortunately superlu-dist has not made a release with the fix yet. We can backport the patch. Since it's changing the entries in colperm_t, I think that makes it an ABI shift, in which case we'd want to add to this transition superlu-dist 8 → 8slu6 Actually no, I think the patch does not change the superlu-dist ABI. Function signatures are the same. The reason I thought it was needed was that petsc continued to segfault after applying the patch. But scrutenising the discussion at superlu-dist issue #141, a second patch is needed, https://github.com/xiaoyeli/superlu_dist/commit/b64fe36742f1468075670129ac460915eb7130fe petsc passes superlu-dist tests after applying the two patches, so I'll upload superlu-dist without an ABI bump.
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
On 2023-07-31 21:30, Sebastian Ramacher wrote: Control: tags -1 confirmed I'd like clear out some transitions in the numerical library stack: combblas: 1.16.0 → 2.0.0 superlu:5 → 6 hypre: 2.26.0 → 2.28.0 mumps:5.5 → 5.6 Please go ahead. All packages are uploaded and built. Unfortunately we're getting test errors from PETSc running its superlu-dist test. The error was actually triggered indirectly by the superlu upgrade. An extra item was inserted into colperm_t in superlu_enum_consts.h. superlu-dist uses it's own copy of the header but hadn't been updated. The discrepancy confuses PETSc so it segfaults. Discussed upstream at superlu-dist issue #141 and PETSc issue #1384. https://github.com/xiaoyeli/superlu_dist/issues/141 https://gitlab.com/petsc/petsc/-/issues/1384 The superlu-dist patch is (I think) https://github.com/xiaoyeli/superlu_dist/commit/f36d65a4e3dee264a18f4bd17b3ec506173ccbc2 Unfortunately superlu-dist has not made a release with the fix yet. We can backport the patch. Since it's changing the entries in colperm_t, I think that makes it an ABI shift, in which case we'd want to add to this transition superlu-dist 8 → 8slu6 Apologies for the inconvenience. Drew
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
On 2023-08-06 20:05, Drew Parsons wrote: hypre test timeouts are an ongoing problem, now chronic on ppc64el. Historically a retry would pass, but ppc64el remains uncooperative. I raised the issue upstream at https://github.com/hypre-space/hypre/issues/955 They suggest making life easier for ourselves and just run `make check`, touching only a subset of all their tests. ppc64el tests continue to hand with make checkpar. We identified that the hypre timeouts occur in pmix, and possibly fixed in pmix v5. Bug#1043263 does not literally block this transition in the sense that I uploaded hypre 2.28.0-6 to just skip ppc64el build tests altogether (can revisit it once pmix v5 is uploaded). With that done, all packages for the transition are uploaded and built, so the transition ready for higher level dependency rebuilds (I'll upload dolfin/dolfinx once petsc is rebuilt). Drew
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
On 2023-08-03 20:46, Adrian Bunk wrote: On Wed, Aug 02, 2023 at 10:58:17AM +0200, Drew Parsons wrote: On 2023-07-31 21:30, Sebastian Ramacher wrote: > > > > combblas: 1.16.0 → 2.0.0 > > superlu:5 → 6 > > hypre: 2.26.0 → 2.28.0 > > mumps:5.5 → 5.6 > > Please go ahead. combblas and superlu are loaded. Probably best to run the rebuild of superlu-dist against combblas ... That didn't build: ... At first sight this looks like a bug in the combblas headers to me. CombBLAS is now fixed, upstream notified https://github.com/PASSIONLab/CombBLAS/issues/20 All main packages are uploaded. mumps has lost powerpc and x32 but we won't stop the transition for that. hypre test timeouts are an ongoing problem, now chronic on ppc64el. Historically a retry would pass, but ppc64el remains uncooperative. I raised the issue upstream at https://github.com/hypre-space/hypre/issues/955 They suggest making life easier for ourselves and just run `make check`, touching only a subset of all their tests. Since these timeouts have always been a problem (I even get them locally on my own system from time to time), we should probably accept their advice. Possibly that will let us reactivate tests for armhf. BTW: dolfin/fenics-basix/fenics-dolfinx need source-only uploads for testing migration. Yes, I figured it's simplest to consider them honorary members of this transition. That is I'll upload them after petsc gets rebuilt for this transition.
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
On 2023-08-04 14:12, Drew Parsons wrote: Oops, BipartiteMatchings is provided twice. I didn't notice that before :( Wait, not as simple as that. It changed location, my dlocate database hasn't caught up yet. Will have to review the patches introduced via https://github.com/xiaoyeli/superlu_dist/issues/60
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
On 2023-08-03 20:46, Adrian Bunk wrote: combblas and superlu are loaded. Probably best to run the rebuild of superlu-dist against combblas ... That didn't build: https://buildd.debian.org/status/logs.php?pkg=superlu-dist=8.1.2%2Bdfsg1-1%2Bb2 ... /usr/bin/ld: CMakeFiles/superlu_dist.dir/z_c2cpp_GetHWPM.cpp.o:/usr/include/CombBLAS/BipartiteMatchings/BPMaximalMatching.h:17: multiple definition of `GlobalMT'; CMakeFiles/superlu_dist.dir/d_c2cpp_GetHWPM.cpp.o:/usr/include/CombBLAS/BipartiteMatchings/BPMaximalMatching.h:17: first defined here /usr/bin/ld: CMakeFiles/superlu_dist.dir/z_c2cpp_GetHWPM.cpp.o: in function `combblas::ThreadBuffLenForBinning(int, int)': /usr/include/CombBLAS/BipartiteMatchings/ApproxWeightPerfectMatching.h:401: multiple definition of `combblas::ThreadBuffLenForBinning(int, int)'; CMakeFiles/superlu_dist.dir/d_c2cpp_GetHWPM.cpp.o:/usr/include/CombBLAS/BipartiteMatchings/ApproxWeightPerfectMatching.h:401: first defined here Oops, BipartiteMatchings is provided twice. I didn't notice that before :(
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
On Wed, Aug 02, 2023 at 10:58:17AM +0200, Drew Parsons wrote: > On 2023-07-31 21:30, Sebastian Ramacher wrote: > > > > > > combblas: 1.16.0 → 2.0.0 > > > superlu:5 → 6 > > > hypre: 2.26.0 → 2.28.0 > > > mumps:5.5 → 5.6 > > > > Please go ahead. > > combblas and superlu are loaded. > > Probably best to run the rebuild of superlu-dist against combblas >... That didn't build: https://buildd.debian.org/status/logs.php?pkg=superlu-dist=8.1.2%2Bdfsg1-1%2Bb2 ... /usr/bin/ld: CMakeFiles/superlu_dist.dir/z_c2cpp_GetHWPM.cpp.o:/usr/include/CombBLAS/BipartiteMatchings/BPMaximalMatching.h:17: multiple definition of `GlobalMT'; CMakeFiles/superlu_dist.dir/d_c2cpp_GetHWPM.cpp.o:/usr/include/CombBLAS/BipartiteMatchings/BPMaximalMatching.h:17: first defined here /usr/bin/ld: CMakeFiles/superlu_dist.dir/z_c2cpp_GetHWPM.cpp.o: in function `combblas::ThreadBuffLenForBinning(int, int)': /usr/include/CombBLAS/BipartiteMatchings/ApproxWeightPerfectMatching.h:401: multiple definition of `combblas::ThreadBuffLenForBinning(int, int)'; CMakeFiles/superlu_dist.dir/d_c2cpp_GetHWPM.cpp.o:/usr/include/CombBLAS/BipartiteMatchings/ApproxWeightPerfectMatching.h:401: first defined here ... At first sight this looks like a bug in the combblas headers to me. > Drew cu Adrian BTW: dolfin/fenics-basix/fenics-dolfinx need source-only uploads for testing migration.
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
On 2023-07-31 21:30, Sebastian Ramacher wrote: combblas: 1.16.0 → 2.0.0 superlu:5 → 6 hypre: 2.26.0 → 2.28.0 mumps:5.5 → 5.6 Please go ahead. combblas and superlu are loaded. Probably best to run the rebuild of superlu-dist against combblas before uploading the new hypre. Drew
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
Control: tags -1 confirmed On 2023-07-25 16:48:02 +0200, Drew Parsons wrote: > Package: release.debian.org > Severity: normal > User: release.debian@packages.debian.org > Usertags: transition > X-Debbugs-Cc: mu...@packages.debian.org, debian-scie...@lists.debian.org > Control: affects -1 + src:mumps > > I'd like clear out some transitions in the numerical library stack: > > combblas: 1.16.0 → 2.0.0 > superlu:5 → 6 > hypre: 2.26.0 → 2.28.0 > mumps:5.5 → 5.6 Please go ahead. Cheers -- Sebastian Ramacher
Bug#1042005: transition: mumps hypre2.28.0 superlu combblas
Package: release.debian.org Severity: normal User: release.debian@packages.debian.org Usertags: transition X-Debbugs-Cc: mu...@packages.debian.org, debian-scie...@lists.debian.org Control: affects -1 + src:mumps I'd like clear out some transitions in the numerical library stack: combblas: 1.16.0 → 2.0.0 superlu:5 → 6 hypre: 2.26.0 → 2.28.0 mumps:5.5 → 5.6 I want to bump hypre to 2.28.0 not 2.29.0 since PETSc cannot yet use hypre 2.29.0. I do not want to bump PETSc to 3.19 yet, because of errors with dolfin (#1042000) and dolfinx (#1042003). Each package has auto-transitions: https://release.debian.org/transitions/html/auto-combblas.html https://release.debian.org/transitions/html/auto-superlu.html https://release.debian.org/transitions/html/auto-hypre.html (albeit for 2.29.0) https://release.debian.org/transitions/html/auto-mumps.html Ben file: title = "hypre"; is_affected = .depends ~ /\b(libhypre\-2\.28\.0|libhypre64\-2\.28\.0|libhypre64m\-2\.28\.0|libhypre\-2\.26\.0|libhypre64\-2\.26\.0|libhypre64m\-2\.26\.0)\b/ is_good = .depends ~ /\b(libhypre\-2\.28\.0|libhypre64\-2\.28\.0|libhypre64m\-2\.28\.0)\b/ is_bad = .depends ~ /\b(libhypre\-2\.26\.0|libhypre64\-2\.26\.0|libhypre64m\-2\.26\.0)\b/ title = "mumps"; is_affected = .depends ~ "libmumps-5.5" | .depends ~ "libmumps-5.6"; is_good = .depends ~ "libmumps-5.6"; is_bad = .depends ~ "libmumps-5.5";
