DOC: Fixes absent line numbers on link to classes decorated with set_module #28918
Conversation
|
@melissawm and @ngoldbaum Would you mind reviewing a follow up from #28645? |
doc/source/conf.py
Outdated
| fn = relpath(fn, start=dirname(numpy.__file__)) | ||
| # This can be removed when removing the decorator set_module. Fix issue #28629 | ||
| if hasattr(obj, '_module_source'): | ||
| obj = copy(obj) |
There was a problem hiding this comment.
this is a no-op:
>>> import copy
>>> class A: ...
...
>>> copy.copy(A) is A
Trueand it's the same story with copy.deepcopy:
>>> copy.deepcopy(A) is A
TrueThere was a problem hiding this comment.
Thanks for pointing that, just got to realize that copy is a little bit subtle.
The intention of the copy was to not introducing any side effect, even while running on the documentation.
Now, the whole thing about changing __module__ is about its side effect and, as discussed previously it was needed downstream.
Since this code is running while generating the docs, do you think we could just recover the original __module__ without further problems?
There was a problem hiding this comment.
Since this code is running while generating the docs, do you think we could just recover the original
__module__without further problems?
I guess that depends on whether the docs currently (also) use the patched __module__ somewhere, and if so, where/when that happens. Put differently; it shouldn't be a problem if it doesn't also change the documented module paths.
But I'm not very familiar with this doc generation code, so it's probably best if another maintainer could chime in here.
There was a problem hiding this comment.
As long as the docs build fine and there aren't any other regressions introduced it should be fine. You could try looking at a diff of the docs build before and after applying the change to see what changes.
It looks like there are some new warnings in the docs build - those need to be fixed too.
There was a problem hiding this comment.
Ok. I will take a look at this and give the feedback here.
There was a problem hiding this comment.
@jorenham and @ngoldbaum.
This is ready for review.
de6f864 to
6ed0ece
Compare
There was a problem hiding this comment.
the poly1d docs now indeed include line numbers in the source code link:
The line numbers are offset because this branch isn't up-to-date with main. Replacing the url and branch with this one, shows that the line numbers are indeed correct: https://github.com/aureliobarbosa/numpy/blob/fix_classes_with_incorrect_line_numbers/numpy/lib/_polynomial_impl.py#L1087-L1457
|
Thanks @aureliobarbosa! Sorry for taking a little while to get back to this. |
|
Thanks @ngoldbaum |
PR #28645 was intended to fix issue #28629, which indicated that wrong links were appearing on documentation for classes decorated with set_module. A minor problem appeared on that fix, since line numbers were not assigned to the link on those decorated classes.
This PR fixes this problem by storing the original value of property
__module__of decorated classes (instead of capturing the property__file__of the original module).In my opinion the current approach is simpler.