fix handling of uninitialized vlen strings#110
Conversation
|
OK, first I've changed the test to have uninitialized values to really see it break. Then, I've added the fix (and some more fixes 😬). Getting this in would help for the h5netcdf integration. I somehow deactivated vlen datasets over there and found it breaking with uninitialized data after reinstating the tests proper. |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #110 +/- ##
==========================================
+ Coverage 72.73% 73.31% +0.57%
==========================================
Files 11 11
Lines 2505 2548 +43
Branches 382 394 +12
==========================================
+ Hits 1822 1868 +46
+ Misses 580 576 -4
- Partials 103 104 +1 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
I've added a similar fix for vlen attributes (found via |
|
@kmuehlbauer cheers - just let me know when this one's good to go in, mate 🍺 |
I'll add a test for the uninitialized attribute for completeness. Hang on some minutes 😁 |
|
I've had to recreate the breaking test from the h5netcdf testsuite. It used some mismatched dimension scales to trigger the error. I've put it in the |
|
Think this is ready from my side, when the lights get green. 💚 |
Description
Reading (partly) uninitialized vlen string datasets breaks because NULL values are not properly handled when fetching from global heap. Also fillvalue's need proper distribution.
Closes #109
Before you get started
Checklist