Back-reference optimizations #272
Conversation
| if full_name in self._searchindex['objects']: | ||
| value = self._searchindex['objects'][full_name] | ||
| if isinstance(value, dict): | ||
| value = value[next(iter(value.keys()))] |
There was a problem hiding this comment.
Note that this is completely wrong. For example, in plot_gallery_version, np.random.random is linked to np.random.lognormal. I believe this is also in part due to #273.
There was a problem hiding this comment.
I just tested this PR. but in this case it does not even recognize np.random.random I'll see with #273
There was a problem hiding this comment.
Definitely should be because of #273.
| else: | ||
| value = int(value) | ||
|
|
||
| docopts[key] = value |
There was a problem hiding this comment.
Can you make a test for this function. Or at least move the parse to a separate function and then test that part at least.
There was a problem hiding this comment.
Test added, but this parsing is a separate function already?
sphinx_gallery/docs_resolv.py
Outdated
| @@ -30,6 +30,9 @@ | |||
|
|
|||
| from io import StringIO | |||
|
|
|||
| from sphinx.search import js_index | |||
| from sphinx.util import jsdump | |||
| if full_name in self._searchindex['objects']: | ||
| value = self._searchindex['objects'][full_name] | ||
| if isinstance(value, dict): | ||
| value = value[next(iter(value.keys()))] |
There was a problem hiding this comment.
I just tested this PR. but in this case it does not even recognize np.random.random I'll see with #273
| else: | ||
| value = int(value) | ||
|
|
||
| docopts[key] = value |
There was a problem hiding this comment.
Can you provide a test for this function. Or at least move this parse to an independent function an test that part at least.
|
this sounds like a great addition, thanks @QuLogic ... I don't have a ton of experience w/ intersphinx code but I agree that it'd be good to test any new parsing functionality built in here. |
|
just a ping that #273 is merged now! |
|
This works well on my machine. I relaunched Circle CI, but it did not make the build on top of master to include #273, so I could not check on the other versions. |
|
On my OSX machine, on current |
This removes a lot of fragile manual parsing that is already maintained elsewhere.
There are only two callers and neither use these options.
This uses the data in the search index exactly as the JavaScript code does, meaning there's no need to search through the HTML of the page to check for the anchor.
Otherwise, it's very vague what URL is failing.
|
Rebased and |
|
Merged. Thanks @QuLogic |
Finding back-references involves downloading a file for every possibly used object in order to do two things:
All this downloading makes back-referencing slow, so instead:
DOCUMENTATION_OPTIONSand find theFILE_SUFFIXfrom there. If it's not defined, it's a really old Sphinx so assume.rst.html.With these two changes, we can stop downloading all those files and just download 2 per back-reference URL. Additionally, use Sphinx's built-in parser to cut down on the parsing code.