TEST: 1.10.x + blas variants#224
Conversation
|
Hi! This is the friendly automated conda-forge-linting service. I just wanted to let you know that I linted all conda-recipes in your PR ( |
Scipy 1.10.0TL;DR: Very similar to 1.9.3, but slight improvementsFrom 32 failures out of 96 runs for 1.9.3, we're now at 28 failures out of 96 CI runs Notable
The good news:
The bad news
Restrictions
Details
** sum of Segfaults (S), Failures (F), resp. Timeouts (T); out of a total of 96 CI combinations being tested Build logs: linux + x64 + mkl + avx512: 1 failurelinux + x64 + openblas + non-avx512: 1 failureosx + x64 + accelerate: 6 failuresosx + x64 + blis: 1 failurewindows + mkl + avx512: 1 failurewindows + openblas: 7 failures |
366e03c to
047adc3
Compare
|
Paused the runs here until we figure out #232. |
66fa60a to
8d8ba75
Compare
Scipy 1.10.1TL;DR: Now with PyPy & PPC, import error on linux+netlib, new failuresFrom 28 failures out of 96 runs for 1.10.0, we're now at 66 failures out of 138 CI runs Notable
The good news:
The bad news
Restrictions
Details
** sum of Segfaults (S), Failures (F), resp. Timeouts (T); out of a total of 138 CI combinations being tested Build logs: (linux or win) + x64 + mkl + avx512: 1 failurelinux + netlib: ImportError! ☢️linux + x64 + openblas + non-avx512: 1 failureosx + x64 + accelerate: 7 failuresosx + x64 + blis: 1 failureosx + x64 + mkl + CPython: 13-14 failuresosx + x64 + mkl + PyPy: 2-3 failureswindows + openblas: 7 failures |
8d8ba75 to
5638bfb
Compare
|
Ah well, turns out my rebase wasn't that great after all: |
|
Seems something about those patches isn't distutils- and/or windows-compatible; but we'll just exclude them on windows for main then... |
Scipy 1.10.1 (distutils)This already ran a few days ago to verify a suspicion that scipy/scipy#18371 had something to do with the switch to meson. While looking at the results of testing PROPRACK in #200, I noticed some familiar failures and ended up writing this up because it's relevant. TL;DR: Switch to meson broke a bunch of thingsFrom 66 failures out of 138 runs for 1.10.1 built with meson (on unix), we're now at 44 failures out of 138 CI runs Notable
Restrictions & Library Versionsas for 1.10.1
** sum of Segfaults (S), Failures (F), resp. Timeouts (T); out of a total of 138 CI combinations being tested Build logs: (linux or win) + x64 + mkl + avx512: 1 failurelinux + x64 + openblas + non-avx512: 1 failureosx + x64 + accelerate: 7 failuresosx + x64 + blis: 1 failureosx + x64 + mkl + pypy: 2-3 failureswindows + openblas: 7 failures |
Scipy 1.10.1 (meson, with patches)Now again with meson after the backported patches from #233 TL;DR: OSX+MKL still broken on mesonFrom 66 failures out of 138 runs for 1.10.1 built with meson (on unix) and 44 failures when built with distutils, we're now at 48 failures out of 138 CI runs Notable
Restrictions & Library Versionsas for 1.10.1
** sum of Segfaults (S), Failures (F), resp. Timeouts (T); out of a total of 138 CI combinations being tested Build logs: (linux or win) + x64 + mkl + avx512: 1 failurelinux + x64 + openblas + non-avx512: 1 failureosx + x64 + accelerate: 7 failuresosx + x64 + blis: 1 failureosx + x64 + mkl + CPython: 13-14 failuresosx + x64 + mkl + pypy: 2-3 failureswindows + openblas: 7 failures |
xref conda-forge/scipy-feedstock#224 for these test failures. The `test_failure_to_run_iterations` issue that is the only failure for 3 OS/BLAS combinations was already fixed in `main`, so with this PR we should be close to most jobs passing again. [skip cirrus] [skip circle]
|
All of the entries with a single failure are from the same test, The precision issues in the macOS/Accelerate tests should be fixed by scipy/scipy#18442. Then there is one left: that is a crash, probably a faulty G77 ABI wrapper. If that happens again after the current round of |
xref conda-forge/scipy-feedstock#224 for these test failures. The `test_failure_to_run_iterations` issue that is the only failure for 3 OS/BLAS combinations was already fixed in `main`, so with this PR we should be close to most jobs passing again. [skip cirrus] [skip circle]
|
Closing this as obsolete |
Following the same scheme as #223, #199, #172, #153 & #130. Should not be merged for the same reasons as #153.