Old system was using manual string formatting for log messages.
This caused issues with common operations like exception logging
because often they need to be handled differently for Py2/Py3
compatibility. In order to unify the effort:
- All logging is changed to `logging.level(msg, arg1, arg2)` style.
- A `SafeLogger` is implemented to auto-decode exceptions properly
in the args (ref #1403).
- Custom formatters were overriding useful logging functionality
like traceback outputing (ref #1402). They are refactored to be
more transparent. Traceback information is provided in `--debug`
mode for `read_file` errors in generators.
- Formatters will now auto-format multiline log messages in order
to make them look related. Similarly, traceback will be formatted in
the same fashion.
- `pelican.log.LimitFilter` was (ab)using logging message which
would result in awkward syntax for argumented logging style. This
functionality is moved to `extra` keyword argument.
- Levels for errors that would result skipping a file (`read_file`)
changed from `warning` to `error` in order to make them stand out
among other logs.
- Small consistency changes to log messages (i.e. changing all
to start with an uppercase letter) and quality-of-life improvements
(some log messages were dumping raw object information).
On OSX, if LC_TIME and LC_CTYPE differs the output of strftime is not properly decoded
in Python 3. This makes sure that the 'utils.DateFormatter' and the related Jinja filter
'strftime' set the same value for LC_TIME and LC_CTYPE while formatting.
Also, '%a' is removed from DEFAULT_DATE_FORMAT in 'custom_locale' tests. OSX and *nix have
different conversions for '%a' ('Jeu' vs 'jeu.') and there is not a feasible way to handle
the difference for tests.
reverts getpelican/pelican@ddcccfeaa9
If one used a locale that made use of unicode characters (like fr_FR.UTF-8)
the files on disk would be in correct locale while links would be to C.
Uses a SafeDatetime class that works with unicode format strigns
by using custom strftime to prevent ascii decoding errors with Python2.
Also added unicode decoding for the calendar module to fix period
archives.
CACHE_PATH can now be relative to settings file like OUTPUT_PATH.
Also add --cache-path commandline option.
Change cache loading warning to a less scary and more helpful message.
This is a reworked and improved version of content caching.
Notable changes:
- by default only raw content and metadata returned by readers are
cached which should prevent conficts with plugins, the speed benefit
of content objects caching is not very big with a simple setup
- renamed --full-rebuild to --ignore-cache
- added more elaborate logging to caching code
The _cache_open attribute of the FileDataCacher class was not set when
settings[load_policy_key] was not True, so saving later failed.
As a precaution, replaced the `if ...: return` style with a plain
if structure to prevent such readability issues and added tests.
`copy('', 'a/b.ext0', 'c/d.ext1')` is copying `a/b.ext0` into `c/d.ext1/b.ext0`
(creating folder `c/d.ext1` in the process) instead of `c/d.ext1`.
Bug introduced by e03cf3f517.
If more than one path is defined in THEME_STATIC_PATHS, the theme's
static directory in output is deleted and replaced by the following
path's files.
Using `shutil.rmtree` to remove the entire destination tree if overwrite
is `True` assumes that we didn't want anything at all that was there. We
should recurse through the directory and their subdirs instead, leaving
things put there by the previous path where they were.
I lazily copied almost verbatim the solution for recursively copying a
diectory from http://stackoverflow.com/a/1994840.
The reason for this is patch is that without it, my plugin is broken! It
also makes my code a lot less crazy:
a83f066
The `slugify()` function used by Pelican is in general very good at
coming up with something both readable and URL-safe. However, there are
a few specific cases where it causes conflicts. One that I've run into
is using the strings `C++` and `C` as tags, both of which transform to
the slug `c`. This commit adds an optional `SLUG_SUBSTITUTIONS` setting
which is a list of 2-tuples of substitutions to be carried out
case-insensitively just prior to stripping out non-alphanumeric
characters. This allows cases like `C++` to be transformed to `CPP` or
similar. This can also improve the readability of slugs.
If DELETE_OUTPUT_DIRECTORY is set to True, all files and directories are
deleted from the output directory. There are, however, several reasons
one might want to retain certain files/directories and avoid their
deletion from the output directory. One such use case is version control
system data: a versioned output directory can facilitate deployment via
Heroku and/or allow the user to easily revert to a prior version of the
site without having to rely on regeneration via Pelican.
This change introduces the OUTPUT_RETENTION setting, a tuple of
filenames that will be preserved when the clean_output_dir function in
pelican.utils is run. Setting OUTPUT_RETENTION = (".hg", ".git") would,
for example, prevent the relevant VCS data from being deleted when the
output directory is cleaned.
Support the forms listed by the W3C [1]. I also removed the
'%Y-%d-%m' form, which can be confused with the '%Y-%m-%d' ISO form.
The new ISO forms can use 'Z' to designate UTC or '[+-]HHMM' to
specify offsets from UTC. Other time zone designators are not
supported.
The '%z' directive has only been supported since Python 3.2 [2], so if
you're running Pelican on Python 2.7, you're stuck with 'Z' for UTC.
Conveniently, we get ValueErrors for both invalid directives and
data/format missmatches, so we don't need special handling for the 2.7
case inside get_date().
[1]: http://www.w3.org/TR/NOTE-datetime
[2]: http://bugs.python.org/issue6641
The old get_relative_path() implementation assumed os.sep == '/',
which doesn't hold on MS Windows. The new implementation uses
split_all() for a more general component count.
I added path_to_url(), because the:
'/'.join(split_all(path))
idiom was showing up in a number of cases, and it's easier to
understand what's going on when that reads:
path_to_url(path)
This will fix a number of places where I think paths and URLs were
conflated, and should improve MS Windows support.
I think the conversion from native paths to URLs is best put off until
we are actually trying to generate the URL. The old handling
(introduced in 2692586, Fixes#645 - Making cross-content linking
windows compatible, 2012-12-19) converted the path at StaticContent
initialization, which left you with a bogus StaticContent.src.
Once we drop the 'static' subdirectory, we will be able to drop the
`dest` and `url` parts from the StaticGenerator.generate_context()
handling, which will leave things looking a good deal cleaner than
they do now.
