Fix pixi update: relax Mantid version pins for Python 3.12#993
Open
fanchercm wants to merge 6 commits into
Open
Fix pixi update: relax Mantid version pins for Python 3.12#993fanchercm wants to merge 6 commits into
fanchercm wants to merge 6 commits into
Conversation
Every Mantid conda channel (mantid and mantid-ornl, main/rc/nightly) has pruned Python 3.11 (6.15.x) builds and now only publishes 6.16.x builds for Python 3.12, breaking the solve for all environments: - Drop the `<6.16.0` upper bound on mantidworkbench in the base, qa, and prod dependencies, which excluded every remaining build in those channels. - Remove the unconstrained `pyqt = "*"` dependency, which pulled in PyQt5 and conflicted with mantidqt's now-mandatory PyQt6 requirement; qtpy remains as the binding-agnostic layer. - Bump requires-python to >=3.12 to match. Verified `pixi update` solves cleanly for default, dev, prod, and qa, and pyrs/qtpy/mantidqt import successfully in each. Findings recorded in docs/ground_truths.md. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Relaxing the mantidworkbench pin (636594c) moved the environment from PyQt5 to PyQt6, which broke two things that were previously masked: - pyrs/icons/icons_rc5.py is a pyrcc-compiled resource module hard-coded to `from PyQt5 import QtCore`, imported by peak_fitting_viewer.py. With PyQt5 no longer installed, this raised ModuleNotFoundError during test collection, aborting the whole pytest session in CI. Replaced the compiled-resource lookup with direct file paths into pyrs/icons and dropped the now-unused icons_rc5.py/icons.qrc. - QTableWidgetItem.setCheckState() is strictly typed in PyQt6 and no longer accepts bool/int; several call sites passed plain bools/ints. Because the collection error above aborted the session first, these paths in fit_table.py and detector_calibration_viewer.py were never actually exercised by CI. Switched them to Qt.CheckState enum values. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
pixi-build-backends is a rolling/nightly channel whose pixi-build-python builds are date+hash suffixed (e.g. 0.5.2.20260612.1208.0827e23) and never match a bare "==0.5.2". Listed ahead of conda-forge (which has a real, clean-versioned 0.5.2 release), it shadowed that match under strict channel priority, making the pin permanently unsolvable and failing both the "Build conda package" and vulnerability-scan build-package CI jobs. Verified by clearing the local rattler cache entirely and rebuilding from cold. Findings recorded in docs/ground_truths.md. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The channel-reorder fix for the conda-build backend didn't actually fix CI: the real cause was that pixi-build-python==0.5.2 only speaks protocol pixi-build-api-version>=4,<5, while prefix-dev/setup-pixi (no pixi-version pin) installs whatever pixi CLI is latest at run time and had already moved to needing >=5,<6. No channel order can satisfy an old exact pin against a newer host CLI. Relaxed the pin to "*" instead. Reproduced CI's exact failure locally by matching its pixi version (0.72.2) with `pixi self-update`, and confirmed the real fix solves from a fully cleared cache. Also, now that the PyQt5 import crash no longer aborts test collection, two more PyQt6 strict-typing bugs surfaced and ran for the first time: - manual_reduction_viewer.py's _mask_state/_calibration_state/_output_state compared a `state` param (Qt.CheckState from checkState(), or plain int from the stateChanged signal) against the bare Qt.Unchecked enum member. PyQt6's Qt.CheckState doesn't compare equal to a plain int in either direction, so the comparison was always wrong for one of the two call paths -- the mask/calibration/output line edits stayed permanently disabled after the user toggled the checkbox, silently keeping the stale default file path. Added an _is_checked() normalizer. Same bare-int-vs-enum bug fixed in fit_table.py's checkState() == 0 check. - strain_stress_view.py's QFormLayout.setFieldGrowthPolicy(0) hit the same strict-enum-typing issue as setCheckState -- switched to QFormLayout.FieldGrowthPolicy.AllNonFixedFieldsGrow. Findings recorded in docs/ground_truths.md. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The tests job passed all 354 tests (0 failures) but still failed CI: ~40s after pytest's summary line, the process segfaulted (exit 139) during interpreter shutdown. This matches a "wrapped C/C++ object of type SliceViewerView has been deleted" RuntimeError seen locally from mantidqt's own ADS-observer cleanup code -- caught and non-fatal under the local offscreen Qt platform, but apparently a genuine use-after-free under CI's real xvfb+xcb display. Added a session-scoped autouse fixture that clears the AnalysisDataService once, after the last test finishes. Pytest fixture teardown runs while the interpreter is still fully alive, well before atexit/shutdown, so any SliceViewer-owned workspaces/observers left open by UI tests are torn down in a safe, well-defined order instead of racing with native object destruction at exit. Verified no regressions: full local suite still passes 354/354 twice in a row. Could not directly reproduce the segfault locally (offscreen apparently doesn't trigger it), so this fix relies on the next CI run for confirmation. Findings recorded in docs/ground_truths.md. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The tests job passed all 354 tests but still failed CI: ~40s after
pytest's own summary line, the process segfaulted (exit 139) during
Python interpreter shutdown, in third-party Mantid/mantidqt native
cleanup code. Clearing the AnalysisDataService at session end (previous
commit) did not fix this -- the exact same crash recurred on the next
CI run, so ADS state wasn't the cause.
Since the crash happens well after every test result and the coverage
report have already been finalized, the fix is to stop the interpreter
from reaching the crashing teardown at all: added
scripts/development/run_tests.py, which calls pytest.main() directly
and then explicitly flushes stdout/stderr and calls os._exit(code) --
skipping the native cleanup that follows. pyproject.toml's `test` pixi
task now runs this wrapper instead of invoking pytest directly.
Two dead ends recorded in docs/ground_truths.md: a pytest_sessionfinish
hook fires before the terminal reporter's own summary and coverage
table are printed (silently truncating the log even though the exit
code was correct), and os._exit() doesn't flush Python's buffered
stdout/stderr the way normal interpreter shutdown does.
Separately: while iterating on the above (which requires running the
full suite locally, matching what CI runs), tests/ui/test_calibration_ui.py
turned out to write/delete files at bare relative paths
("HB2B_test_export.json", "HB2B_CAL.json") -- i.e. in whatever the
current working directory happens to be, which is the repo root for
both CI and a local run. This collided with and destroyed pre-existing
untracked files of the same name in the repo root. Fixed the test to
write under pytest's tmp_path fixture instead, so it can never touch a
developer's working directory regardless of what's sitting there.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
mantidandmantid-ornl, main/rc/nightly) has pruned Python 3.11 (6.15.x) builds and now only publishes 6.16.x builds for Python 3.12, breakingpixi update/pixi installon this branch.<6.16.0upper bound onmantidworkbenchin the base,qa, andproddependencies — it was excluding every remaining build in those channels.pyqt = "*"dependency, which pulled in PyQt5 and conflicted withmantidqt's now-mandatory PyQt6 requirement;qtpyremains as the binding-agnostic layer.requires-pythonto>=3.12to match.docs/ground_truths.mdfor future reference.Test plan
pixi updatesolves cleanly fordefault,dev,prod, andqaenvironments (previously failed with a solver conflict).pixi run -e default test-import-frameworkpasses (importspyrs,qtpy,mantidqt).pixi run -e dev test-import-frameworkpasses.pixi run -e prod test-import-frameworkpasses.🤖 Generated with Claude Code