Fix slicing bug in ensure_memoryview#6449
Conversation
|
should ultimately update the comment above this code too |
|
@JacksonMaxfield, would you be able to try this out? 🙂 |
|
In class, gimme an hour or so and will do! |
|
No rush. Thanks! 🙏 |
Unit Test Results 15 files ± 0 15 suites ±0 6h 19m 34s ⏱️ - 1m 15s For more details on these failures, see this check. Results for commit e1ccade. ± Comparison against base commit 046ab17. ♻️ This comment has been updated with latest results. |
|
@jakirkham here are the checks for this branch install: https://github.com/AllenCellModeling/aicsimageio/runs/6600770595?check_suite_focus=true The important tests are the Cannot believe this is less than a single line change to make it work. Fun. |
|
That said, maybe wait for macOS builds to finish up but your branch is working so far. |
|
Thanks for checking Jackson! 🙏 Glad to hear it works 😄 |
|
Trying to reproduce the issue myself so I can come up with a test. Tried to do But ran into some issues. Is there something else I need to do to download this file? Edit: NVM think I figured this out |
|
Hah, yea sorry that is just the viewer and UI downloader. |
|
Yeah sorry poked around a bit and found the right URL. Now have it reproducing for me locally. Going to dig into it a bit more |
This causes problems with certain kinds of views. So just pass the `memoryview` as-is even though `PickleBuffer` will chain the underlying `memoryview`s instead of referencing the original `obj`. Maybe this can be fixed upstream.
|
Ok found a test that covers a similar problem. Think it is likely the cause of the issue in the original repro, but it is a little difficult for me to parse fully atm. In any event, this test will fail without the change here, which should ensure we don't regress |
db6653f to
d7c56f9
Compare
This reproduces a similar issue to what is seen in the original repro. Most importantly it fails if `mv.obj` is used.
|
rerun tests Note: rerunning gpuCI tests since those errors should be fixed by #6434 |
|
Restarted the failed GitHub Actions for good measure. Edit: Though they looked like flaky tests. |
jrbourbeau
left a comment
There was a problem hiding this comment.
Thanks for the quick fix @jakirkham and reporting the original issue @JacksonMaxfield. This LGTM, will merge once CI finishes
|
All of the tests passed except for the macOS test, which is still queued. Do we want to wait for that? |
|
The previous failure for that build was a flaky test, so happy to merge in |
|
Thanks James! Also thanks Jackson for testing 🙏 |
Fixes #6448
The underlying issues seems to be related to slicing. This example demonstrates the issue below and has been included in the test suite.
pre-commit run --all-files