BUG: revert trim_zeros changes from gh-16911 #17171
Conversation
| def test_list_to_list(self): | ||
| res = trim_zeros(self.a.tolist()) | ||
| assert isinstance(res, list) |
There was a problem hiding this comment.
I assume this test does not pass on master?
There was a problem hiding this comment.
It does as self.a is just a normal 1D array of integers.
However, there is currently no test which checks the return type so I feel this is a nice addition.
There was a problem hiding this comment.
I'm trying to work out whether this is a case of @mattip generally wishing to reseal the can of worms, or if there was a specific worm that we knew was an incompatibility.
There was a problem hiding this comment.
Mainly seal the can of worms. I am a bit dubious if the new implementation actually achieves what it promises, personally. (Depending on the use-case, you may be stripping only one value from a very large array, making the old code much faster. OTOH, maybe that use-case doesn't actually matter since in almost all cases the work done on the result will be much slower anyway.)
If you are fairly certain the updated version seals it well, I think we can go with that as well. I had never looked very carefully, but it seemed a bit hard to be sure previously, and trim_zeros seemed a bit niche to worry about much...
There was a problem hiding this comment.
The PR reverts the code to its state before the first PR gh-16911. I added the passing test to cover a use case that previously was not tested.
|
@eric-wieser We also want it resolved quickly so that the weekly wheels get built for testing. |
|
Then lets put this in for now/ At least for the wheels... Sorry if you want to still pursue the initial improvement, which will be a bit more difficult with this reverting. Thanks for all the effort @BvB93 ! |
|
Ah well, it was worth a shot. |
xref PRs gh-16911, gh-17058, gh-17131 and issue gh-16783.
While the effort to speed up and rethink
trim_zerosis appreciated, in this case I think we should unfortunately revert these changes. I have added a test that lists remain unchanged.