This is the repository for the online module Bots as Digital Infrapuncture, commissioned by the Utrecht University
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

982 lines
37 KiB

import datetime
import locale
import logging
import os.path
from posixpath import join as posix_join
from sys import platform
from jinja2.utils import generate_lorem_ipsum
from pelican.contents import Article, Author, Category, Page, Static
from pelican.plugins.signals import content_object_init
from pelican.settings import DEFAULT_CONFIG
from pelican.tests.support import (LoggedTestCase, get_context, get_settings,
unittest)
from pelican.utils import (path_to_url, posixize_path, truncate_html_words)
# generate one paragraph, enclosed with <p>
TEST_CONTENT = str(generate_lorem_ipsum(n=1))
TEST_SUMMARY = generate_lorem_ipsum(n=1, html=False)
class TestBase(LoggedTestCase):
def setUp(self):
super().setUp()
self.old_locale = locale.setlocale(locale.LC_ALL)
locale.setlocale(locale.LC_ALL, 'C')
self.page_kwargs = {
'content': TEST_CONTENT,
'context': {
'localsiteurl': '',
},
'metadata': {
'summary': TEST_SUMMARY,
'title': 'foo bar',
'author': Author('Blogger', DEFAULT_CONFIG),
},
'source_path': '/path/to/file/foo.ext'
}
self._disable_limit_filter()
def tearDown(self):
locale.setlocale(locale.LC_ALL, self.old_locale)
self._enable_limit_filter()
def _disable_limit_filter(self):
from pelican.contents import logger
logger.disable_filter()
def _enable_limit_filter(self):
from pelican.contents import logger
logger.enable_filter()
def _copy_page_kwargs(self):
# make a deep copy of page_kwargs
page_kwargs = {key: self.page_kwargs[key] for key in self.page_kwargs}
for key in page_kwargs:
if not isinstance(page_kwargs[key], dict):
break
page_kwargs[key] = {
subkey: page_kwargs[key][subkey] for subkey in page_kwargs[key]
}
return page_kwargs
class TestPage(TestBase):
def test_use_args(self):
# Creating a page with arguments passed to the constructor should use
# them to initialise object's attributes.
metadata = {'foo': 'bar', 'foobar': 'baz', 'title': 'foobar', }
page = Page(TEST_CONTENT, metadata=metadata,
context={'localsiteurl': ''})
for key, value in metadata.items():
self.assertTrue(hasattr(page, key))
self.assertEqual(value, getattr(page, key))
self.assertEqual(page.content, TEST_CONTENT)
def test_mandatory_properties(self):
# If the title is not set, must throw an exception.
page = Page('content')
self.assertFalse(page._has_valid_mandatory_properties())
self.assertLogCountEqual(
count=1,
msg="Skipping .*: could not find information about 'title'",
level=logging.ERROR)
page = Page('content', metadata={'title': 'foobar'})
self.assertTrue(page._has_valid_mandatory_properties())
def test_summary_from_metadata(self):
# If a :summary: metadata is given, it should be used
page = Page(**self.page_kwargs)
self.assertEqual(page.summary, TEST_SUMMARY)
def test_summary_max_length(self):
# If a :SUMMARY_MAX_LENGTH: is set, and there is no other summary,
# generated summary should not exceed the given length.
page_kwargs = self._copy_page_kwargs()
settings = get_settings()
page_kwargs['settings'] = settings
del page_kwargs['metadata']['summary']
settings['SUMMARY_MAX_LENGTH'] = None
page = Page(**page_kwargs)
self.assertEqual(page.summary, TEST_CONTENT)
settings['SUMMARY_MAX_LENGTH'] = 10
page = Page(**page_kwargs)
self.assertEqual(page.summary, truncate_html_words(TEST_CONTENT, 10))
settings['SUMMARY_MAX_LENGTH'] = 0
page = Page(**page_kwargs)
self.assertEqual(page.summary, '')
def test_summary_end_suffix(self):
# If a :SUMMARY_END_SUFFIX: is set, and there is no other summary,
# generated summary should contain the specified marker at the end.
page_kwargs = self._copy_page_kwargs()
settings = get_settings()
page_kwargs['settings'] = settings
del page_kwargs['metadata']['summary']
settings['SUMMARY_END_SUFFIX'] = 'test_marker'
settings['SUMMARY_MAX_LENGTH'] = 10
page = Page(**page_kwargs)
self.assertEqual(page.summary, truncate_html_words(TEST_CONTENT, 10,
'test_marker'))
self.assertIn('test_marker', page.summary)
def test_summary_get_summary_warning(self):
"""calling ._get_summary() should issue a warning"""
page_kwargs = self._copy_page_kwargs()
page = Page(**page_kwargs)
self.assertEqual(page.summary, TEST_SUMMARY)
self.assertEqual(page._get_summary(), TEST_SUMMARY)
self.assertLogCountEqual(
count=1,
msg=r"_get_summary\(\) has been deprecated since 3\.6\.4\. "
"Use the summary decorator instead",
level=logging.WARNING)
def test_slug(self):
page_kwargs = self._copy_page_kwargs()
settings = get_settings()
page_kwargs['settings'] = settings
settings['SLUGIFY_SOURCE'] = "title"
page = Page(**page_kwargs)
self.assertEqual(page.slug, 'foo-bar')
settings['SLUGIFY_SOURCE'] = "basename"
page = Page(**page_kwargs)
self.assertEqual(page.slug, 'foo')
# test slug from title with unicode and case
inputs = (
# (title, expected, preserve_case, use_unicode)
('指導書', 'zhi-dao-shu', False, False),
('指導書', 'Zhi-Dao-Shu', True, False),
('指導書', '指導書', False, True),
('指導書', '指導書', True, True),
('Çığ', 'cig', False, False),
('Çığ', 'Cig', True, False),
('Çığ', 'çığ', False, True),
('Çığ', 'Çığ', True, True),
)
settings = get_settings()
page_kwargs = self._copy_page_kwargs()
page_kwargs['settings'] = settings
for title, expected, preserve_case, use_unicode in inputs:
settings['SLUGIFY_PRESERVE_CASE'] = preserve_case
settings['SLUGIFY_USE_UNICODE'] = use_unicode
page_kwargs['metadata']['title'] = title
page = Page(**page_kwargs)
self.assertEqual(page.slug, expected,
(title, preserve_case, use_unicode))
def test_defaultlang(self):
# If no lang is given, default to the default one.
page = Page(**self.page_kwargs)
self.assertEqual(page.lang, DEFAULT_CONFIG['DEFAULT_LANG'])
# it is possible to specify the lang in the metadata infos
self.page_kwargs['metadata'].update({'lang': 'fr', })
page = Page(**self.page_kwargs)
self.assertEqual(page.lang, 'fr')
def test_save_as(self):
# If a lang is not the default lang, save_as should be set
# accordingly.
# if a title is defined, save_as should be set
page = Page(**self.page_kwargs)
self.assertEqual(page.save_as, "pages/foo-bar.html")
# if a language is defined, save_as should include it accordingly
self.page_kwargs['metadata'].update({'lang': 'fr', })
page = Page(**self.page_kwargs)
self.assertEqual(page.save_as, "pages/foo-bar-fr.html")
def test_relative_source_path(self):
# 'relative_source_path' should be the relative path
# from 'PATH' to 'source_path'
page_kwargs = self._copy_page_kwargs()
# If 'source_path' is None, 'relative_source_path' should
# also return None
page_kwargs['source_path'] = None
page = Page(**page_kwargs)
self.assertIsNone(page.relative_source_path)
page_kwargs = self._copy_page_kwargs()
settings = get_settings()
full_path = page_kwargs['source_path']
settings['PATH'] = os.path.dirname(full_path)
page_kwargs['settings'] = settings
page = Page(**page_kwargs)
# if 'source_path' is set, 'relative_source_path' should
# return the relative path from 'PATH' to 'source_path'
self.assertEqual(
page.relative_source_path,
os.path.relpath(
full_path,
os.path.dirname(full_path)
))
def test_metadata_url_format(self):
# Arbitrary metadata should be passed through url_format()
page = Page(**self.page_kwargs)
self.assertIn('summary', page.url_format.keys())
page.metadata['directory'] = 'test-dir'
page.settings = get_settings(PAGE_SAVE_AS='{directory}/{slug}')
self.assertEqual(page.save_as, 'test-dir/foo-bar')
def test_datetime(self):
# If DATETIME is set to a tuple, it should be used to override LOCALE
dt = datetime.datetime(2015, 9, 13)
page_kwargs = self._copy_page_kwargs()
# set its date to dt
page_kwargs['metadata']['date'] = dt
page = Page(**page_kwargs)
# page.locale_date is a unicode string in both python2 and python3
dt_date = dt.strftime(DEFAULT_CONFIG['DEFAULT_DATE_FORMAT'])
self.assertEqual(page.locale_date, dt_date)
page_kwargs['settings'] = get_settings()
# I doubt this can work on all platforms ...
if platform == "win32":
locale = 'jpn'
else:
locale = 'ja_JP.utf8'
page_kwargs['settings']['DATE_FORMATS'] = {'jp': (locale,
'%Y-%m-%d(%a)')}
page_kwargs['metadata']['lang'] = 'jp'
import locale as locale_module
try:
page = Page(**page_kwargs)
self.assertEqual(page.locale_date, '2015-09-13(\u65e5)')
except locale_module.Error:
# The constructor of ``Page`` will try to set the locale to
# ``ja_JP.utf8``. But this attempt will failed when there is no
# such locale in the system. You can see which locales there are
# in your system with ``locale -a`` command.
#
# Until we find some other method to test this functionality, we
# will simply skip this test.
unittest.skip("There is no locale %s in this system." % locale)
def test_template(self):
# Pages default to page, metadata overwrites
default_page = Page(**self.page_kwargs)
self.assertEqual('page', default_page.template)
page_kwargs = self._copy_page_kwargs()
page_kwargs['metadata']['template'] = 'custom'
custom_page = Page(**page_kwargs)
self.assertEqual('custom', custom_page.template)
def test_signal(self):
def receiver_test_function(sender):
receiver_test_function.has_been_called = True
pass
receiver_test_function.has_been_called = False
content_object_init.connect(receiver_test_function)
self.assertIn(
receiver_test_function,
content_object_init.receivers_for(Page))
self.assertFalse(receiver_test_function.has_been_called)
Page(**self.page_kwargs)
self.assertTrue(receiver_test_function.has_been_called)
def test_get_content(self):
# Test that the content is updated with the relative links to
# filenames, tags and categories.
settings = get_settings()
args = self.page_kwargs.copy()
args['settings'] = settings
# Tag
args['content'] = ('A simple test, with a '
'<a href="|tag|tagname">link</a>')
page = Page(**args)
content = page.get_content('http://notmyidea.org')
self.assertEqual(
content,
('A simple test, with a '
'<a href="http://notmyidea.org/tag/tagname.html">link</a>'))
# Category
args['content'] = ('A simple test, with a '
'<a href="|category|category">link</a>')
page = Page(**args)
content = page.get_content('http://notmyidea.org')
self.assertEqual(
content,
('A simple test, with a '
'<a href="http://notmyidea.org/category/category.html">link</a>'))
def test_intrasite_link(self):
cls_name = '_DummyArticle'
article = type(cls_name, (object,), {'url': 'article.html'})
args = self.page_kwargs.copy()
args['settings'] = get_settings()
args['source_path'] = 'content'
args['context']['generated_content'] = {'article.rst': article}
# Classic intrasite link via filename
args['content'] = (
'A simple test, with a '
'<a href="|filename|article.rst">link</a>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'A simple test, with a '
'<a href="http://notmyidea.org/article.html">link</a>'
)
# fragment
args['content'] = (
'A simple test, with a '
'<a href="|filename|article.rst#section-2">link</a>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'A simple test, with a '
'<a href="http://notmyidea.org/article.html#section-2">link</a>'
)
# query
args['content'] = (
'A simple test, with a '
'<a href="|filename|article.rst'
'?utm_whatever=234&highlight=word">link</a>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'A simple test, with a '
'<a href="http://notmyidea.org/article.html'
'?utm_whatever=234&highlight=word">link</a>'
)
# combination
args['content'] = (
'A simple test, with a '
'<a href="|filename|article.rst'
'?utm_whatever=234&highlight=word#section-2">link</a>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'A simple test, with a '
'<a href="http://notmyidea.org/article.html'
'?utm_whatever=234&highlight=word#section-2">link</a>'
)
# also test for summary in metadata
parsed = (
'A simple summary test, with a '
'<a href="|filename|article.rst">link</a>'
)
linked = (
'A simple summary test, with a '
'<a href="http://notmyidea.org/article.html">link</a>'
)
args['settings']['FORMATTED_FIELDS'] = ['summary', 'custom']
args['metadata']['summary'] = parsed
args['metadata']['custom'] = parsed
args['context']['localsiteurl'] = 'http://notmyidea.org'
p = Page(**args)
# This is called implicitly from all generators and Pelican.run() once
# all files are processed. Here we process just one page so it needs
# to be called explicitly.
p.refresh_metadata_intersite_links()
self.assertEqual(p.summary, linked)
self.assertEqual(p.custom, linked)
def test_intrasite_link_more(self):
cls_name = '_DummyAsset'
args = self.page_kwargs.copy()
args['settings'] = get_settings()
args['source_path'] = 'content'
args['context']['static_content'] = {
'images/poster.jpg':
type(cls_name, (object,), {'url': 'images/poster.jpg'}),
'assets/video.mp4':
type(cls_name, (object,), {'url': 'assets/video.mp4'}),
'images/graph.svg':
type(cls_name, (object,), {'url': 'images/graph.svg'}),
}
args['context']['generated_content'] = {
'reference.rst':
type(cls_name, (object,), {'url': 'reference.html'}),
}
# video.poster
args['content'] = (
'There is a video with poster '
'<video controls poster="{static}/images/poster.jpg">'
'<source src="|static|/assets/video.mp4" type="video/mp4">'
'</video>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'There is a video with poster '
'<video controls poster="http://notmyidea.org/images/poster.jpg">'
'<source src="http://notmyidea.org/assets/video.mp4"'
' type="video/mp4">'
'</video>'
)
# object.data
args['content'] = (
'There is a svg object '
'<object data="{static}/images/graph.svg"'
' type="image/svg+xml">'
'</object>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'There is a svg object '
'<object data="http://notmyidea.org/images/graph.svg"'
' type="image/svg+xml">'
'</object>'
)
# blockquote.cite
args['content'] = (
'There is a blockquote with cite attribute '
'<blockquote cite="{filename}reference.rst">blah blah</blockquote>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'There is a blockquote with cite attribute '
'<blockquote cite="http://notmyidea.org/reference.html">'
'blah blah'
'</blockquote>'
)
def test_intrasite_link_absolute(self):
"""Test that absolute URLs are merged properly."""
args = self.page_kwargs.copy()
args['settings'] = get_settings(
STATIC_URL='http://static.cool.site/{path}',
ARTICLE_URL='http://blog.cool.site/{slug}.html')
args['source_path'] = 'content'
args['context']['static_content'] = {
'images/poster.jpg':
Static('', settings=args['settings'],
source_path='images/poster.jpg'),
}
args['context']['generated_content'] = {
'article.rst':
Article('', settings=args['settings'], metadata={
'slug': 'article', 'title': 'Article'})
}
# Article link will go to blog
args['content'] = (
'<a href="{filename}article.rst">Article</a>'
)
content = Page(**args).get_content('http://cool.site')
self.assertEqual(
content,
'<a href="http://blog.cool.site/article.html">Article</a>'
)
# Page link will go to the main site
args['content'] = (
'<a href="{index}">Index</a>'
)
content = Page(**args).get_content('http://cool.site')
self.assertEqual(
content,
'<a href="http://cool.site/index.html">Index</a>'
)
# Image link will go to static
args['content'] = (
'<img src="{static}/images/poster.jpg"/>'
)
content = Page(**args).get_content('http://cool.site')
self.assertEqual(
content,
'<img src="http://static.cool.site/images/poster.jpg"/>'
)
def test_intrasite_link_markdown_spaces(self):
cls_name = '_DummyArticle'
article = type(cls_name, (object,), {'url': 'article-spaces.html'})
args = self.page_kwargs.copy()
args['settings'] = get_settings()
args['source_path'] = 'content'
args['context']['generated_content'] = {'article spaces.rst': article}
# An intrasite link via filename with %20 as a space
args['content'] = (
'A simple test, with a '
'<a href="|filename|article%20spaces.rst">link</a>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'A simple test, with a '
'<a href="http://notmyidea.org/article-spaces.html">link</a>'
)
def test_intrasite_link_source_and_generated(self):
"""Test linking both to the source and the generated article
"""
cls_name = '_DummyAsset'
args = self.page_kwargs.copy()
args['settings'] = get_settings()
args['source_path'] = 'content'
args['context']['generated_content'] = {
'article.rst': type(cls_name, (object,), {'url': 'article.html'})}
args['context']['static_content'] = {
'article.rst': type(cls_name, (object,), {'url': 'article.rst'})}
args['content'] = (
'A simple test, with a link to an'
'<a href="{filename}article.rst">article</a> and its'
'<a href="{static}article.rst">source</a>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'A simple test, with a link to an'
'<a href="http://notmyidea.org/article.html">article</a> and its'
'<a href="http://notmyidea.org/article.rst">source</a>'
)
def test_intrasite_link_to_static_content_with_filename(self):
"""Test linking to a static resource with deprecated {filename}
"""
cls_name = '_DummyAsset'
args = self.page_kwargs.copy()
args['settings'] = get_settings()
args['source_path'] = 'content'
args['context']['static_content'] = {
'poster.jpg':
type(cls_name, (object,), {'url': 'images/poster.jpg'})}
args['content'] = (
'A simple test, with a link to a'
'<a href="{filename}poster.jpg">poster</a>'
)
content = Page(**args).get_content('http://notmyidea.org')
self.assertEqual(
content,
'A simple test, with a link to a'
'<a href="http://notmyidea.org/images/poster.jpg">poster</a>'
)
def test_multiple_authors(self):
"""Test article with multiple authors."""
args = self.page_kwargs.copy()
content = Page(**args)
assert content.authors == [content.author]
args['metadata'].pop('author')
args['metadata']['authors'] = [Author('First Author', DEFAULT_CONFIG),
Author('Second Author', DEFAULT_CONFIG)]
content = Page(**args)
assert content.authors
assert content.author == content.authors[0]
class TestArticle(TestBase):
def test_template(self):
# Articles default to article, metadata overwrites
default_article = Article(**self.page_kwargs)
self.assertEqual('article', default_article.template)
article_kwargs = self._copy_page_kwargs()
article_kwargs['metadata']['template'] = 'custom'
custom_article = Article(**article_kwargs)
self.assertEqual('custom', custom_article.template)
def test_slugify_category_author(self):
settings = get_settings()
settings['SLUG_REGEX_SUBSTITUTIONS'] = [
(r'C#', 'csharp'),
(r'[^\w\s-]', ''),
(r'(?u)\A\s*', ''),
(r'(?u)\s*\Z', ''),
(r'[-\s]+', '-'),
]
settings['ARTICLE_URL'] = '{author}/{category}/{slug}/'
settings['ARTICLE_SAVE_AS'] = '{author}/{category}/{slug}/index.html'
article_kwargs = self._copy_page_kwargs()
article_kwargs['metadata']['author'] = Author("O'Brien", settings)
article_kwargs['metadata']['category'] = Category(
'C# & stuff', settings)
article_kwargs['metadata']['title'] = 'fnord'
article_kwargs['settings'] = settings
article = Article(**article_kwargs)
self.assertEqual(article.url, 'obrien/csharp-stuff/fnord/')
self.assertEqual(
article.save_as, 'obrien/csharp-stuff/fnord/index.html')
def test_slugify_with_author_substitutions(self):
settings = get_settings()
settings['AUTHOR_REGEX_SUBSTITUTIONS'] = [
('Alexander Todorov', 'atodorov'),
('Krasimir Tsonev', 'krasimir'),
(r'[^\w\s-]', ''),
(r'(?u)\A\s*', ''),
(r'(?u)\s*\Z', ''),
(r'[-\s]+', '-'),
]
settings['ARTICLE_URL'] = 'blog/{author}/{slug}/'
settings['ARTICLE_SAVE_AS'] = 'blog/{author}/{slug}/index.html'
article_kwargs = self._copy_page_kwargs()
article_kwargs['metadata']['author'] = Author('Alexander Todorov',
settings)
article_kwargs['metadata']['title'] = 'fnord'
article_kwargs['settings'] = settings
article = Article(**article_kwargs)
self.assertEqual(article.url, 'blog/atodorov/fnord/')
self.assertEqual(article.save_as, 'blog/atodorov/fnord/index.html')
def test_slugify_category_with_dots(self):
settings = get_settings()
settings['CATEGORY_REGEX_SUBSTITUTIONS'] = [
('Fedora QA', 'fedora.qa'),
]
settings['ARTICLE_URL'] = '{category}/{slug}/'
article_kwargs = self._copy_page_kwargs()
article_kwargs['metadata']['category'] = Category('Fedora QA',
settings)
article_kwargs['metadata']['title'] = 'This Week in Fedora QA'
article_kwargs['settings'] = settings
article = Article(**article_kwargs)
self.assertEqual(article.url, 'fedora.qa/this-week-in-fedora-qa/')
def test_valid_save_as_detects_breakout(self):
settings = get_settings()
article_kwargs = self._copy_page_kwargs()
article_kwargs['metadata']['slug'] = '../foo'
article_kwargs['settings'] = settings
article = Article(**article_kwargs)
self.assertFalse(article._has_valid_save_as())
def test_valid_save_as_detects_breakout_to_root(self):
settings = get_settings()
article_kwargs = self._copy_page_kwargs()
article_kwargs['metadata']['slug'] = '/foo'
article_kwargs['settings'] = settings
article = Article(**article_kwargs)
self.assertFalse(article._has_valid_save_as())
def test_valid_save_as_passes_valid(self):
settings = get_settings()
article_kwargs = self._copy_page_kwargs()
article_kwargs['metadata']['slug'] = 'foo'
article_kwargs['settings'] = settings
article = Article(**article_kwargs)
self.assertTrue(article._has_valid_save_as())
class TestStatic(LoggedTestCase):
def setUp(self):
super().setUp()
self.settings = get_settings(
STATIC_SAVE_AS='{path}',
STATIC_URL='{path}',
PAGE_SAVE_AS=os.path.join('outpages', '{slug}.html'),
PAGE_URL='outpages/{slug}.html')
self.context = get_context(self.settings)
self.static = Static(content=None, metadata={}, settings=self.settings,
source_path=posix_join('dir', 'foo.jpg'),
context=self.context)
self.context['static_content'][self.static.source_path] = self.static
def tearDown(self):
pass
def test_attach_to_same_dir(self):
"""attach_to() overrides a static file's save_as and url.
"""
page = Page(
content="fake page",
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'fakepage.md'))
self.static.attach_to(page)
expected_save_as = os.path.join('outpages', 'foo.jpg')
self.assertEqual(self.static.save_as, expected_save_as)
self.assertEqual(self.static.url, path_to_url(expected_save_as))
def test_attach_to_parent_dir(self):
"""attach_to() preserves dirs inside the linking document dir.
"""
page = Page(content="fake page", metadata={'title': 'fakepage'},
settings=self.settings, source_path='fakepage.md')
self.static.attach_to(page)
expected_save_as = os.path.join('outpages', 'dir', 'foo.jpg')
self.assertEqual(self.static.save_as, expected_save_as)
self.assertEqual(self.static.url, path_to_url(expected_save_as))
def test_attach_to_other_dir(self):
"""attach_to() ignores dirs outside the linking document dir.
"""
page = Page(content="fake page",
metadata={'title': 'fakepage'}, settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'))
self.static.attach_to(page)
expected_save_as = os.path.join('outpages', 'foo.jpg')
self.assertEqual(self.static.save_as, expected_save_as)
self.assertEqual(self.static.url, path_to_url(expected_save_as))
def test_attach_to_ignores_subsequent_calls(self):
"""attach_to() does nothing when called a second time.
"""
page = Page(content="fake page",
metadata={'title': 'fakepage'}, settings=self.settings,
source_path=os.path.join('dir', 'fakepage.md'))
self.static.attach_to(page)
otherdir_settings = self.settings.copy()
otherdir_settings.update(dict(
PAGE_SAVE_AS=os.path.join('otherpages', '{slug}.html'),
PAGE_URL='otherpages/{slug}.html'))
otherdir_page = Page(
content="other page",
metadata={'title': 'otherpage'},
settings=otherdir_settings,
source_path=os.path.join('dir', 'otherpage.md'))
self.static.attach_to(otherdir_page)
otherdir_save_as = os.path.join('otherpages', 'foo.jpg')
self.assertNotEqual(self.static.save_as, otherdir_save_as)
self.assertNotEqual(self.static.url, path_to_url(otherdir_save_as))
def test_attach_to_does_nothing_after_save_as_referenced(self):
"""attach_to() does nothing if the save_as was already referenced.
(For example, by a {static} link an a document processed earlier.)
"""
original_save_as = self.static.save_as
page = Page(
content="fake page",
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'fakepage.md'))
self.static.attach_to(page)
self.assertEqual(self.static.save_as, original_save_as)
self.assertEqual(self.static.url, path_to_url(original_save_as))
def test_attach_to_does_nothing_after_url_referenced(self):
"""attach_to() does nothing if the url was already referenced.
(For example, by a {static} link an a document processed earlier.)
"""
original_url = self.static.url
page = Page(
content="fake page",
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'fakepage.md'))
self.static.attach_to(page)
self.assertEqual(self.static.save_as, self.static.source_path)
self.assertEqual(self.static.url, original_url)
def test_attach_to_does_not_override_an_override(self):
"""attach_to() does not override paths that were overridden elsewhere.
(For example, by the user with EXTRA_PATH_METADATA)
"""
customstatic = Static(
content=None,
metadata=dict(save_as='customfoo.jpg', url='customfoo.jpg'),
settings=self.settings,
source_path=os.path.join('dir', 'foo.jpg'),
context=self.settings.copy())
page = Page(
content="fake page",
metadata={'title': 'fakepage'}, settings=self.settings,
source_path=os.path.join('dir', 'fakepage.md'))
customstatic.attach_to(page)
self.assertEqual(customstatic.save_as, 'customfoo.jpg')
self.assertEqual(customstatic.url, 'customfoo.jpg')
def test_attach_link_syntax(self):
"""{attach} link syntax triggers output path override & url replacement.
"""
html = '<a href="{attach}../foo.jpg">link</a>'
page = Page(
content=html,
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'),
context=self.context)
content = page.get_content('')
self.assertNotEqual(
content, html,
"{attach} link syntax did not trigger URL replacement.")
expected_save_as = os.path.join('outpages', 'foo.jpg')
self.assertEqual(self.static.save_as, expected_save_as)
self.assertEqual(self.static.url, path_to_url(expected_save_as))
def test_tag_link_syntax(self):
"{tag} link syntax triggers url replacement."
html = '<a href="{tag}foo">link</a>'
page = Page(
content=html,
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'),
context=self.context)
content = page.get_content('')
self.assertNotEqual(content, html)
def test_category_link_syntax(self):
"{category} link syntax triggers url replacement."
html = '<a href="{category}foo">link</a>'
page = Page(
content=html,
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'),
context=self.context)
content = page.get_content('')
self.assertNotEqual(content, html)
def test_author_link_syntax(self):
"{author} link syntax triggers url replacement."
html = '<a href="{author}foo">link</a>'
page = Page(
content=html,
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'),
context=self.context)
content = page.get_content('')
self.assertNotEqual(content, html)
def test_index_link_syntax(self):
"{index} link syntax triggers url replacement."
html = '<a href="{index}">link</a>'
page = Page(
content=html,
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'),
context=self.context)
content = page.get_content('')
self.assertNotEqual(content, html)
expected_html = ('<a href="' +
'/'.join((self.settings['SITEURL'],
self.settings['INDEX_SAVE_AS'])) +
'">link</a>')
self.assertEqual(content, expected_html)
def test_unknown_link_syntax(self):
"{unknown} link syntax should trigger warning."
html = '<a href="{unknown}foo">link</a>'
page = Page(content=html,
metadata={'title': 'fakepage'}, settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'),
context=self.context)
content = page.get_content('')
self.assertEqual(content, html)
self.assertLogCountEqual(
count=1,
msg="Replacement Indicator 'unknown' not recognized, "
"skipping replacement",
level=logging.WARNING)
def test_link_to_unknown_file(self):
"{filename} link to unknown file should trigger warning."
html = '<a href="{filename}foo">link</a>'
page = Page(content=html,
metadata={'title': 'fakepage'}, settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'),
context=self.context)
content = page.get_content('')
self.assertEqual(content, html)
self.assertLogCountEqual(
count=1,
msg="Unable to find 'foo', skipping url replacement.",
level=logging.WARNING)
def test_index_link_syntax_with_spaces(self):
"""{index} link syntax triggers url replacement
with spaces around the equal sign."""
html = '<a href = "{index}">link</a>'
page = Page(
content=html,
metadata={'title': 'fakepage'},
settings=self.settings,
source_path=os.path.join('dir', 'otherdir', 'fakepage.md'),
context=self.context)
content = page.get_content('')
self.assertNotEqual(content, html)
expected_html = ('<a href = "' +
'/'.join((self.settings['SITEURL'],
self.settings['INDEX_SAVE_AS'])) +
'">link</a>')
self.assertEqual(content, expected_html)
def test_not_save_as_draft(self):
"""Static.save_as is not affected by draft status."""
static = Static(
content=None,
metadata=dict(status='draft',),
settings=self.settings,
source_path=os.path.join('dir', 'foo.jpg'),
context=self.settings.copy())
expected_save_as = posixize_path(os.path.join('dir', 'foo.jpg'))
self.assertEqual(static.status, 'draft')
self.assertEqual(static.save_as, expected_save_as)
self.assertEqual(static.url, path_to_url(expected_save_as))