REF: de-duplicate DST tzconversion code #35077
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
also rebase |
|
Pretty much every variant of this I can come up with segfaults on Linux but not on Mac. cc @chris-b1 any thoughts? |
|
Any chance you've tried to debug the segfault using the steps outlined in #35100? Might be helpful to step through the debugger (replace lldb with gdb for linux most likely) |
|
@jbrockmendel which test(s) are segfaulting? I don't see any in CI |
WillAyd
reviewed
Jul 7, 2020
They're not segfaulting ATM because they are being caught by debugging assertions first. e.g. https://dev.azure.com/pandas-dev/pandas/_build/results?buildId=38537&view=logs&j=a3a13ea8-7cf0-5bdb-71bb-6ac8830ae35c&t=add65f64-6c25-5783-8fd6-d9aa1b63d9d4&l=160 the assertion that raises is https://github.com/pandas-dev/pandas/pull/35077/files?file-filters%5B%5D=.pxd&file-filters%5B%5D=.pyx#diff-03511a52390e3160aa5d428a7b45a4f4R57, but if we get to that line, then we should have already executed https://github.com/pandas-dev/pandas/pull/35077/files?file-filters%5B%5D=.pxd&file-filters%5B%5D=.pyx#diff-4260fb485bcfabb1da9cc5a85ed6d639R265, which isnt raising despite asserting the same thing. Locally I just got a segfault inside tests.frame.test_to_csv, but running the tests on just that file generally doesnt produce the segfault (well, 2 runs out of the last 5) |
|
This seems to confirm the suspicion that this is a |
|
My latest thought is that the |
|
^ expanding on the above, if within info.positions = <intp_t*>cnp.PyArray_DATA(pos)to from libc.string cimport memcpy
from cpython.mem cimport PyMem_Malloc
info.positions = <intp_t *>PyMem_Malloc(n * sizeof(intp_t))
memcpy(info.positions, pos.data, n * sizeof(intp_t))Then the tests pass. This is probably not the best way to code this and leaks memory as is, but I think confirms the suspicion that memory management in |
This seems right, good call. In a different branch where I use a cdef class instead of a struct, retaining the |
WillAyd
reviewed
Jul 7, 2020
| int noffsets | ||
| int64_t* utcoffsets | ||
| intp_t* positions | ||
| ndarray positions_arr # needed to avoid segfault |
There was a problem hiding this comment.
Do we even need both positions_arr and positions or can we just use the former?
There was a problem hiding this comment.
we could just use the former, but i think we get a perf boost from indexing on the latter
There was a problem hiding this comment.
If you can see a difference then sure, but if not would definitely be cleaner to just stick with the ndarray
|
ATM getting a non-trivial perf hit; I think I didnt see this with the struct version |
|
If I can't get the perf figured out here, we can at least get some mileage de-duplication-wise by sharing+inlining some helpers following #35168 |
|
If the struct version is faster you could also memcpy the bytes needed; would just need a cleanup function to free them whenever done with the struct |
|
The struct version is a little bit faster, but still a non-trivial perf hit: |
|
Closing, will revisit following #35168 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This implements TZConvertInfo to de-duplicate a bunch of tzconversion code. This uses the new pattern on get_resolution. The other usages I intend to update in individual follow-ups in which I'll check that we have good asv coverage the affected functions.
Sits on top of #35075 (actually, that was broken off of this)
Perf-improving according to the newly implemented asvs:
L50-L57 in libresolution is going to end up being repeated, should become a helper function. Doing that separately since it tentatively looks like that may have a perf hit if not done carefully.
xref 652a3de#r30437889