Drop code target, sphinx-autogen is not using conf.py, and since Sphinx
4.0 autosummary is generating stub pages by default anyway.
Mock ctypes and linkcheck.__main__: colorama.py imports MS Windows
libraries, autosummary executes __main__.py.
Default is ASCII.
xgettext: Non-ASCII string at ../linkcheck/htmlutil/srcsetparse.py:39.
Please specify the source encoding through --from-code or through a comment
as specified in http://www.python.org/peps/pep-0263.html.
When checking local files with AnchorCheck, anchors in URLs
like "example/#anchor" are not supported.
Without AnchorCheck enabled, the Real URL reported for such URLs
was changed to include the anchor when local file checking was added to
AnchorCheck, but it is the directory that is checked.
The same URL was also then used as the Parent URL for the check of each
of the contents of that directory.
For FileUrl this is a revert of:
c221afda ("Enable AnchorCheck to be used with local files", 2022-10-03)
I started with a test of urlencoded anchors, assuming at the URL might
have a urlencoded anchor, but the actual anchor in the HTML would NOT be
urlencoded.
"I added a test for file:// processing, and it was showing different
results for when the URL anchor was and wasn't quoted. I tracked it down
to code in fileurl.py that was calling url_norm, and I'm pretty sure the
code is unnecessary at this point. But I made a minimally-invasive
change, to be as safe as possible."
UrlBase.build_url() in line 174 also calls url_norm()
The threading issue has been there for years, but I didn't notice it
until after I thought I was done, while I was doing manual testing
(with threads re-enabled).
The problem was with storing URL-specific state (.anchors) on the
AnchorCheck object itself, because there's only one global AnchorCheck
object, so all the threads are competing to use that one simgle variable
(self.anchors).
The solution was to create a new object to hold .anchors, for each
processed URL.
[I] discovered that fileurl.py was stripping the anchors from url_data,
which breaks AnchorCheck. So I stopped it from doing that, and
tried to fix up all the places that were assuming the url would map to a
filesystem file. The tests all pass, but I'm not 100% sure I caught all
the cases, or fixed them correctly.
