Repository navigation
BUG: Fix buffer export dtype references - #17753
Conversation
|
This needs to be backported if the other buffer fixes get backported. |
There was a problem hiding this comment.
descr is not used until line 554. Could you move the descr = PyArray_DESCR(arr); Py_INCREF(descr); to there instead, or comment why it needs to stay at line 489?
There was a problem hiding this comment.
Yeah, much better, thanks, definitely moved the wrong line around (moved it above now as well, so its at the end of the block in both branches).
|
Could you xref the other PRs that this relates to, so when/if backporting them this has less chance of being forgotten? |
10af778 to
9bbc32e
Compare
|
The PR where this was introduced is gh-17295. |
There is a small refcount bug introduced when fixing/improving the buffer export code recently. This fixes it.
9bbc32e to
a970510
Compare
I'm planning on a 1.19.5 for the segfault fix, and maybe an OpenBLAS update if that gets sorted. 1.20.0 is close enough that I think lesser fixes can wait. How much importance do you put on the buffer fixes? |
|
@charris true, I pinged the two people who opened the issue and regression that they addressed. Lets say, if they reply that it would help them soon enough, we backport :). |
|
I assume you didn't merge because the test run didn't finish. It seems to be handing, so I will just ignore it. Only the refcheck tests were missing it appears, those are really unlikely to be related to anything here. Thanks Matti. |
There is a small refcount bug introduced when fixing/improving the
buffer export code recently. This fixes it.
It would be fine to just merge this, I am running the test more, so I might find more small issues like this, but it might also take another day until I even look/get the test results...