Metadata applied to a directory will apply to all files under
it. In case of conflicts, child paths beat parent paths, so metadata
applied to `dir/subdir/file.md` will take precedence over that applied
to `dir/subdir`, which will take precedence over just `dir`.
This PR removes the Python 3.4 tox task and updates references in the
code to Python 3.5+.
tox complains about Python 3.4, which is EOL after next month:
> py34 installed: DEPRECATION: Python 3.4 support has been deprecated. pip 19.1 will be the last one supporting it. Please upgrade your Python as Python 3.4 won't be maintained after March 2019 (cf PEP 429).
Reverts back to how pagination worked for the {url} placeholder as I did
it in 182fb11c80. Absolute URLs with
one or two leading slashes were eaten by lstrip() and became relative,
which then caused broken links in my case.
Added extra comments to this piece of code (*and* the test) to make it
less likely that someone breaks this again in the future.
3a0add4b6e caused existing configs to fall
back to defaults. But since we know exactly how to fix the user config
so that the behavior doesn't change, we should do so, while still
warning that use of %s is deprecated.
Also fixes a bug where we tried to look for %s in None.
At the moment aborts with the following:
File "/usr/lib/python3.7/site-packages/pelican/settings.py", line 327, in handle_deprecated_settings
if t not in settings['PAGINATED_TEMPLATES']:
KeyError: 'PAGINATED_TEMPLATES'
The refresh_metadata_intersite_links() is called again later, so this
bit was only causing everything to be processed twice. Besides
that, it makes the processing dependent on file order -- in particular,
when metadata references a file that was not parsed yet, it reported an
"Unable to find" warning. But everything is found in the second pass, so
this only causes a superflous false warning and no change to the output.
The related test now needs to call the
refresh_metadata_intersite_links() explicitly. That function is called
from Pelican.run() and all generators but the test processes just one
page so it has no chance of being called implicitly.
Related discussion: https://github.com/getpelican/pelican/pull/2288/files#r204337359
With the current implementation, links to articles/pages that have not
been processed yet are reported with a warning, yet later replaced with
a correct value. That's not wanted.
