Conversation
|
Ok. I think that shows the same bug, according to discussion in #208 |
|
Oh dear. |
|
It has been fixed in pytho-eccodes but not been released on conda yet, only pip. |
Well I'm a bit confused by that. The releases available via conda seem to be long after the merge of the referenced issue ? Meantime, I'm removing this from the 0.16 effort. |
pp-mo
added
Status: Blocked
T-Shirt: Small
Type: Technical Debt
The fix (ecmwf/eccodes-python#31) went in at the end of July. The last conda release of python eccodes was at the end of June, THe last pip release was in August. https://github.com/ecmwf/eccodes-python/releases |
|
The segmentation fault issue with float32 is fixed in the next release of eccodes-python (v1.0.0) which will be accompanied by ecCodes v2.19.0 |
|
Re-fixed and re-targetted at v0.16.x It's now really simple, so let's hope it checks out.. |
1 similar comment
|
Thanks @pp-mo ! |
* upstream/v0.16.x: Docstest 0v16 (SciTools#244) Remove eccodes bug workaround added in SciTools#208. (SciTools#224) Update requirements to pick up Iris 3. (SciTools#243) Add 'main' conda-forge channel, needed for docs builds. (SciTools#240) Fix RC date in release notes (about to cut). (SciTools#235) Version 0.16 release candidate (SciTools#232) Cosmetic change : rename the travis iris-test-version options (SciTools#234) Travis test with both Iris latest-release and latest-master. (SciTools#231)
* Travis test with both Iris latest-release and latest-master. (#231) * Travis test with both Iris latest-release and latest-master. * Modify test CMLs for latest Iris (Iris3.0 changes). * Grib1 load fixes. * Fix loading since units=None default for Iris3 coords * Modify test to work with latest Iris (Iris3.0 changes). * Test against latest Iris only. * Review changes. * Cosmetic change : rename the travis iris-test-version options (#234) * Rename Iris test-version options, and enable all to check action. * Review changes. * Version 0.16 release candidate (#232) * Require Iris >=3 (just for the gdt90 changes). * Whatsnew entry for requiring Iris 3. * Set version string for release candidate. * Test against Iris 3.0.0rc0 from conda-forge/rc_iris. * Fix RC date in release notes (about to cut). (#235) * Add 'main' conda-forge channel, needed for docs builds. (#240) * Add 'main' conda-forge channel, needed for docs builds. * Use fixed spherical-earth-radius in GRIB1, ignoring change in gribapi default. * Codestyle fix. * Update requirements to pick up Iris 3. (#243) * Remove eccodes bug workaround added in #208. (#224) * Docstest 0v16 (#244) * Document PR#240 in release notes. * Fix version string and release-notes date. * Added getting started .cirrus.yml * Update cirrus to use miniconda image * Update .cirrus.yml * Update .cirrus.yml Copied across cirrus file from iris-ugrid * Added nox testing Borrowing from iris and iris-ugrid, added nox testing and a test runner. Tests currently fail on my local machine. * Path fixes in cirrus.yml * eccodes test added to noxfile * Added config and coverage * Trying to set SITE_CFG * syntax error * taking the IRIS_DIR from Travis CI config * Add allow_failures to the linux task for now * Dodgy r key... * Update .cirrus.yml * Allow failures in linting * Force return error 0 for now * Added eccodes test * moved matrix * Yaml syntax error & * wrong way around * Typo * Update iris version dependency * Date fix This resolves the same bash issue as SciTools/iris#4019 * Configure Iris in nox * Updated CI config * correct site-packages path * Fix to iris test data path * force cache update * refactored writing iris config in noxfile * Fix pep8 and license test fails * Support for testing against packaged iris and building from source * Adding yaml to list of cirrus container requirements * invalidate cirrus cache * really invalidate cirrus cache... * yaml -> pyyaml * missing iris_dir reference * call write_iris_config * docstrings and matrix testing * git syntax error * eccodes selfcheck in basic tests * Removed py3.8 testing for now * Fixed regression in setup.py test * Removed unused imports * Disabled lint checking * Removed nox from test dependencies for python3.6 and python3.8 * Removed python3.8 from noxfile * Removed nox from test requirements Co-authored-by: Patrick Peglar <patrick.peglar@metoffice.gov.uk>
5 participants
WIP
Aiming to remove a workaround introduced in #208, supposedly now fixed in python-eccodes.
So far, have re-checked that the original problem is replicable, having forced the old python-eccodes version, and removed the fix. This triggered a segfault in the
tests.integration.round_trip.test_grid_definition_section.TestGDT5. It should also hit this intests/unit/save_rules/test_data_section.TestNonDoubleData(but that is not run before the crash happens).Then, removed the version pin, now testing with 2.18.0 and python-eccodes 2020.06.0.
This should have been ok, but at present the original crash is still occurring .
Current Status : skipped out the GDT5 test, temporarily, to enable the intended test
tests/unit/save_rules/test_data_section.TestNonDoubleData, which targets/demonstrates the problem more clearly.