In 10.8 the bug has been fixed good and proper, by https://github.com/sagemath/sage/pull/40579 (see also https://github.com/sagemath/sage/pull/41325)
Indeed, as William correctly wrote, it had to do with an improper interaction with linbox. On Thursday, December 18, 2025 at 6:25:33 PM UTC-6 Dima Pasechnik wrote: > On Thu, Dec 18, 2025 at 3:29 PM Volker Braun <[email protected]> wrote: > > > > In April i accidentally rewrote 10.6.rc1 version commit from 8a8453f35f3 > to 10741006a47, changing only metadata. That was the only mistake that I'm > aware of. But it only means that 8a8453f35f3 isn't part of the "release > tree". > > > > If you keep going to the first parent commit starting at 10.7 > (85c8f1e8a26) then you end up at 10.6 (b8f98e7c7c3). So 10.7 is most > certainly based on 10.6 in the git sense. > > > > You are probably tripping over messy merges in-between. To bisect you > need --first-parent to only bisect at the release merges. > > > > $ git checkout 10.7 > > $ git bisect start --first-parent > > $ git bisect new HEAD > > $ git bisect old 10.6 > > Bisecting: 229 revisions left to test after this (roughly 8 steps) > > [581aae7712a34b2a143d4e8decc03344ff862aa3] gh-40164: ⬆️ Bump > astral-sh/setup-uv from 6.0.1 to 6.1.0 > > > > Thanks for the tip. The bugfix happened at commit > 3531a873beb5df16d1172525013ba9159f3f84d0, > that is, when https://github.com/sagemath/sage/pull/39733 was merged. > > Basically, it switches the default linear algebra echelonize() method in > src/sage/matrix/matrix_rational_dense.pyx to a different algorithm, > avoiding > the use of "multimodular", i.e. _echelonize_multimodular(), calling > matrix_rational_echelon_form_multimodular() - which apparently does > work correctly with `Parallelism().set(nproc=2)` (or bigger than 2). > > So a bug is still there, it's just hidden, in a way. > > Dima > > > > On Thursday, December 18, 2025 at 9:52:38 PM UTC+1 Dima Pasechnik wrote: > >> > >> Maxim Kontsevich reported patently wrong answers from modular forms > >> code in https://github.com/sagemath/sage/issues/41267. > >> We were able to pin them down to setting Parallelism().set(nproc=k), > >> for any k>1. The error is not dependent upon the platform (observed > >> in Linux Conda originally, but meanwhile found to occur in "normal" > >> builds, too, on Linux x86_64 and on Intel macOS) - arm64 etc still > >> needs to be checked. > >> > >> It would be great to understand what fixed it - any ideas? > >> > >> For reasons unclear to me, the git history between tags 10.6 and 10.7 > >> is not clean (somehow, 10.7 is not "based" upon 10.6 in Git sense), > >> breaking a straightforward git bisect. > >> Help with the latter would be appreciated, too. > >> (otherwise one would need to do a manual git rebase of 10.7 over 10.6, > >> which isn't instant) > >> > >> Dima > >> > >> PS. Computations done in Sage 9.7-10.6 under Parallelism().set(nproc=k) > >> (e.g. one might have set "Parallelism().set(nproc=42)" in > ~/.sage/init.sage/) > >> thus might be incorrect :-( > > > > -- > > You received this message because you are subscribed to the Google > Groups "sage-release" group. > > To unsubscribe from this group and stop receiving emails from it, send > an email to [email protected]. > > To view this discussion visit > https://groups.google.com/d/msgid/sage-release/5e473ce1-29be-4f1e-93e7-d7e2c94c0935n%40googlegroups.com > . > -- You received this message because you are subscribed to the Google Groups "sage-support" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/sage-support/bb098d61-fd6b-4c7b-b051-0f3f4e652524n%40googlegroups.com.
