Fix random VCR failures in Pixiv tests.
Sometimes tests randomly fail because the PHPSESSID they use in their
HTTP requests to Pixiv is different than the one that was originally
recorded by VCR. This causes VCR to complain that the requests don't
match.
This is caused by the PHPSESSID being globally cached in Memcache.
Depending on the order the tests run in (which is random), one set of
tests can use a PHPSESSID that was recorded for a /different/ set of
tests.
Improve Pixiv URL matching.
* Allow URLs that are missing the http:// part. These are sometimes seen
in artist entries.
* Ignore URLs from random Pixiv domains such as dic.pixiv.net,
blog.pixiv.net, etc. These are also sometimes in artist entries.
Improve normalize_for_artist_finder! URL matching.
* Normalize www.pixiv.net/stacc/username URLs.
* Correctly normalize URLs that are missing the illust ID part on the end
(i.e. http://i2.pixiv.net/img04/img/syounen_no_uta/). These are common
in artist entries.
Match URLs strictly when normalizing for artist entries.
Only normalize Pixiv URLs that strictly match a known format. Pass any
unrecognized URLs through without attempting to normalize them, just to
be safe.
Normalize URLs when saving artist entries.
This refactors things such that
* Fetching the ugoira frame data happens at the end of rewrite, not
rewrite_html_pages. This way uploading zip URLs should work.
* The source data is cached in an instance variable in case it
was indeed fetched during rewrite_html_pages. This way it doesn't get
fetched a second time.
file.get_input_stream leaves the file open by default, which means the
zip cannot be moved later on. But the block version of it closes the
file automatically.
mkvmerge didn't realize it was supposed to output a webm specifically
because write_path does not have a file extension. The --webm option
tells it this instead.
Some browsers like Chrome (and some video players like MPC-HC) do not
display the last frame of a webm (or they display it, but only very
briefly) if set to loop. This commit is a hack to force the last frame
to display for the correct amount of time by duplicating it.
* Fix regex for getting frame data not working if agent is logged in to
pixiv
* Fix #unpack not returning the unpacked folder
* Fix frame_data being inaccessible because it was a local variable
* Fix images getting put in tmpdir, when they're expected to be in
tmpdir/images
* Fix deleting the tmpdir manually (it gets deleted automatically)
* Fix output webm being put in tmpdir (which gets deleted) instead of
write_path
