mkdocs-material/material/plugins/blog/plugin.py

896 lines
33 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# Copyright (c) 2016-2023 Martin Donath <martin.donath@squidfunk.com>
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to
# deal in the Software without restriction, including without limitation the
# rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
# sell copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
# IN THE SOFTWARE.
import logging
import os
import paginate
import posixpath
import re
import readtime
import sys
from babel.dates import format_date
from copy import copy
from datetime import date, datetime, time
from hashlib import sha1
from lxml.html import fragment_fromstring, tostring
from mkdocs import utils
from mkdocs.utils.meta import get_data
from mkdocs.commands.build import _populate_page
from mkdocs.contrib.search import SearchIndex
from mkdocs.plugins import BasePlugin
from mkdocs.structure.files import File, Files
from mkdocs.structure.nav import Link, Section
from mkdocs.structure.pages import Page
from tempfile import gettempdir
from yaml import SafeLoader, load
from material.plugins.blog.config import BlogConfig
# -----------------------------------------------------------------------------
# Class
# -----------------------------------------------------------------------------
# Blog plugin
class BlogPlugin(BasePlugin[BlogConfig]):
supports_multiple_instances = True
# Initialize plugin
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# Initialize variables for incremental builds
self.is_serve = False
self.is_dirtyreload = False
self.is_dirty = False
# Determine whether we're running under dirty reload
def on_startup(self, *, command, dirty):
self.is_serve = (command == "serve")
self.is_dirty = dirty
# Initialize plugin
def on_config(self, config):
if not self.config.enabled:
return
# Resolve source directory for posts and generated files
self.post_dir = self._resolve("posts")
self.temp_dir = gettempdir()
# Initialize posts
self.post_map = dict()
self.post_meta_map = dict()
self.post_pages = []
self.post_pager_pages = []
# Initialize archive
if self.config.archive:
self.archive_map = dict()
self.archive_post_map = dict()
# Initialize categories
if self.config.categories:
self.category_map = dict()
self.category_name_map = dict()
self.category_post_map = dict()
# Initialize authors
if self.config.authors:
self.authors_map = dict()
# Resolve authors file
path = os.path.normpath(os.path.join(
config.docs_dir,
self.config.authors_file.format(
blog = self.config.blog_dir
)
))
# Load authors map, if it exists
if os.path.isfile(path):
with open(path, encoding = "utf-8") as f:
self.authors_map = load(f, SafeLoader) or {}
# Ensure that format strings have no trailing slashes
for option in [
"post_url_format",
"archive_url_format",
"categories_url_format",
"pagination_url_format"
]:
if self.config[option].endswith("/"):
log.error(f"Option '{option}' must not contain trailing slash.")
sys.exit(1)
# Inherit global table of contents setting
if not isinstance(self.config.archive_toc, bool):
self.config.archive_toc = self.config.blog_toc
if not isinstance(self.config.categories_toc, bool):
self.config.categories_toc = self.config.blog_toc
# If pagination should not be used, set to large value
if not self.config.pagination:
self.config.pagination_per_page = 1e7
# By default, drafts are rendered when the documentation is served,
# but not when it is built. This should nicely align with the expected
# user experience when authoring documentation.
if self.is_serve and self.config.draft_on_serve:
self.config.draft = True
# Adjust paths to assets in the posts directory and preprocess posts
def on_files(self, files, *, config):
if not self.config.enabled:
return
# Adjust destination paths for assets
path = self._resolve("assets")
for file in files.media_files():
if self.post_dir not in file.src_uri:
continue
# Compute destination URL
file.url = file.url.replace(self.post_dir, path)
# Compute destination file system path
file.dest_uri = file.dest_uri.replace(self.post_dir, path)
file.abs_dest_path = os.path.join(config.site_dir, file.dest_path)
# Hack: as post URLs are dynamically computed and can be configured by
# the author, we need to compute them before we process the contents of
# any other page or post. If we wouldn't do that, URLs would be invalid
# and we would need to patch them afterwards. The only way to do this
# correctly is to first extract the metadata of all posts. Additionally,
# while we're at it, generate all archive and category pages as we have
# the post metadata on our hands. This ensures that we can safely link
# from anywhere to all pages that are generated as part of the blog.
for file in files.documentation_pages():
if self.post_dir not in file.src_uri:
continue
# Read and preprocess post
with open(file.abs_src_path, encoding = "utf-8") as f:
markdown, meta = get_data(f.read())
# Ensure post has a date set
if not meta.get("date"):
log.error(f"Blog post '{file.src_uri}' has no date set.")
sys.exit(1)
# Compute slug from metadata, content or file name
headline = utils.get_markdown_title(markdown)
slug = meta.get("title", headline or file.name)
# Front matter can be defind in YAML, guarded by two lines with
# `---` markers, or MultiMarkdown, separated by an empty line.
# If the author chooses to use MultiMarkdown syntax, date is
# returned as a string, which is different from YAML behavior,
# which returns a date. Thus, we must check for its type, and
# parse the date for normalization purposes.
if isinstance(meta["date"], str):
meta["date"] = date.fromisoformat(meta["date"])
# Normalize date to datetime for proper sorting
if not isinstance(meta["date"], datetime):
meta["date"] = datetime.combine(meta["date"], time())
# Compute category slugs
categories = []
for name in meta.get("categories", []):
categories.append(self.config.categories_slugify(
name, self.config.categories_slugify_separator
))
# Check if maximum number of categories is reached
max_categories = self.config.post_url_max_categories
if len(categories) == max_categories:
break
# Compute path from format string
date_format = self.config.post_url_date_format
path = self.config.post_url_format.format(
categories = "/".join(categories),
date = self._format_date(meta["date"], date_format, config),
file = file.name,
slug = meta.get("slug", self.config.post_slugify(
slug, self.config.post_slugify_separator
))
)
# Normalize path, as it may begin with a slash
path = posixpath.normpath("/".join([".", path]))
# Compute destination URL according to settings
file.url = self._resolve(path)
if not config.use_directory_urls:
file.url += ".html"
else:
file.url += "/"
# Compute destination file system path
file.dest_uri = re.sub(r"(?<=\/)$", "index.html", file.url)
file.abs_dest_path = os.path.join(
config.site_dir, file.dest_path
)
# Add post metadata
self.post_meta_map[file.src_uri] = meta
# Sort post metadata by date (descending)
self.post_meta_map = dict(sorted(
self.post_meta_map.items(),
key = lambda item: item[1]["date"], reverse = True
))
# Find and extract the section hosting the blog
path = self._resolve("index.md")
root = _host(config.nav, path)
# Ensure blog root exists
file = files.get_file_from_path(path)
if not file:
log.error(f"Blog root '{path}' does not exist.")
sys.exit(1)
# Ensure blog root is part of navigation
if not root:
log.error(f"Blog root '{path}' not in navigation.")
sys.exit(1)
# Generate and register files for archive
if self.config.archive:
name = self._translate(config, self.config.archive_name)
data = self._generate_files_for_archive(config, files)
if data:
root.append({ name: data })
# Generate and register files for categories
if self.config.categories:
name = self._translate(config, self.config.categories_name)
data = self._generate_files_for_categories(config, files)
if data:
root.append({ name: data })
# Hack: add posts temporarily, so MkDocs doesn't complain
name = sha1(path.encode("utf-8")).hexdigest()
root.append({
f"__posts_${name}": list(self.post_meta_map.keys())
})
# Cleanup navigation before proceeding
def on_nav(self, nav, *, config, files):
if not self.config.enabled:
return
# Find and resolve index for cleanup
path = self._resolve("index.md")
file = files.get_file_from_path(path)
# Determine blog root section
self.main = file.page
if self.main.parent:
root = self.main.parent.children
else:
root = nav.items
# Hack: remove temporarily added posts from the navigation
name = sha1(path.encode("utf-8")).hexdigest()
for item in root:
if not item.is_section or item.title != f"__posts_${name}":
continue
# Detach previous and next links of posts
if item.children:
head = item.children[+0]
tail = item.children[-1]
# Link page prior to posts to page after posts
if head.previous_page:
head.previous_page.next_page = tail.next_page
# Link page after posts to page prior to posts
if tail.next_page:
tail.next_page.previous_page = head.previous_page
# Contain previous and next links inside posts
head.previous_page = None
tail.next_page = None
# Set blog as parent page
for page in item.children:
page.parent = self.main
next = page.next_page
# Switch previous and next links
page.next_page = page.previous_page
page.previous_page = next
# Remove posts from navigation
root.remove(item)
break
# Prepare post for rendering
def on_page_markdown(self, markdown, *, page, config, files):
if not self.config.enabled:
return
# Only process posts
if self.post_dir not in page.file.src_uri:
return
# Skip processing of drafts
if self._is_draft(page.file.src_uri):
return
# Ensure template is set or use default
if "template" not in page.meta:
page.meta["template"] = "blog-post.html"
# Use previously normalized date
page.meta["date"] = self.post_meta_map[page.file.src_uri]["date"]
# Ensure navigation is hidden
page.meta["hide"] = page.meta.get("hide", [])
if "navigation" not in page.meta["hide"]:
page.meta["hide"].append("navigation")
# Format date for rendering
date_format = self.config.post_date_format
page.meta["date_format"] = self._format_date(
page.meta["date"], date_format, config
)
# Compute readtime if desired and not explicitly set
if self.config.post_readtime:
# There's a bug in the readtime library, which causes it to fail
# when the input string contains emojis (reported in #5555)
encoded = markdown.encode("unicode_escape")
if "readtime" not in page.meta:
rate = self.config.post_readtime_words_per_minute
read = readtime.of_markdown(encoded, rate)
page.meta["readtime"] = read.minutes
# Compute post categories
page.categories = []
if self.config.categories:
for name in page.meta.get("categories", []):
file = files.get_file_from_path(self.category_name_map[name])
page.categories.append(file.page)
# Compute post authors
page.authors = []
if self.config.authors:
for name in page.meta.get("authors", []):
if name not in self.authors_map:
log.error(
f"Blog post '{page.file.src_uri}' author '{name}' "
f"unknown, not listed in .authors.yml"
)
sys.exit(1)
# Add author to page
page.authors.append(self.authors_map[name])
# Fix stale link if previous post is a draft
prev = page.previous_page
while prev and self._is_draft(prev.file.src_uri):
page.previous_page = prev.previous_page
prev = prev.previous_page
# Fix stale link if next post is a draft
next = page.next_page
while next and self._is_draft(next.file.src_uri):
page.next_page = next.next_page
next = next.next_page
# Filter posts and generate excerpts for generated pages
def on_env(self, env, *, config, files):
if not self.config.enabled:
return
# Skip post excerpts on dirty reload to save time
if self.is_dirtyreload:
return
# Copy configuration and enable 'toc' extension
config = copy(config)
config.mdx_configs["toc"] = copy(config.mdx_configs.get("toc", {}))
# Ensure that post titles are links
config.mdx_configs["toc"]["anchorlink"] = True
config.mdx_configs["toc"]["permalink"] = False
# Filter posts that should not be published
for file in files.documentation_pages():
if self.post_dir in file.src_uri:
if self._is_draft(file.src_uri):
files.remove(file)
# Ensure template is set
if "template" not in self.main.meta:
self.main.meta["template"] = "blog.html"
# Populate archive
if self.config.archive:
for path in self.archive_map:
self.archive_post_map[path] = []
# Generate post excerpts for archive
base = files.get_file_from_path(path)
for file in self.archive_map[path]:
self.archive_post_map[path].append(
self._generate_excerpt(file, base, config, files)
)
# Ensure template is set
page = base.page
if "template" not in page.meta:
page.meta["template"] = "blog-archive.html"
# Populate categories
if self.config.categories:
for path in self.category_map:
self.category_post_map[path] = []
# Generate post excerpts for categories
base = files.get_file_from_path(path)
for file in self.category_map[path]:
self.category_post_map[path].append(
self._generate_excerpt(file, base, config, files)
)
# Ensure template is set
page = base.page
if "template" not in page.meta:
page.meta["template"] = "blog-category.html"
# Resolve path of initial index
curr = self._resolve("index.md")
base = self.main.file
# Initialize index
self.post_map[curr] = []
self.post_pager_pages.append(self.main)
# Generate indexes by paginating through posts
for path in self.post_meta_map.keys():
file = files.get_file_from_path(path)
if not self._is_draft(path):
self.post_pages.append(file.page)
else:
continue
# Generate new index when the current is full
per_page = self.config.pagination_per_page
if len(self.post_map[curr]) == per_page:
offset = 1 + len(self.post_map)
# Resolve path of new index
curr = self.config.pagination_url_format.format(page = offset)
curr = self._resolve(curr + ".md")
# Generate file
self._generate_file(curr, f"# {self.main.title}")
# Register file and page
base = self._register_file(curr, config, files)
page = self._register_page(base, config, files)
# Inherit page metadata, title and position
page.meta = self.main.meta
page.title = self.main.title
page.parent = self.main
page.previous_page = self.main.previous_page
page.next_page = self.main.next_page
# Initialize next index
self.post_map[curr] = []
self.post_pager_pages.append(page)
# Assign post excerpt to current index
self.post_map[curr].append(
self._generate_excerpt(file, base, config, files)
)
# Populate generated pages
def on_page_context(self, context, *, page, config, nav):
if not self.config.enabled:
return
# Provide post excerpts for index
path = page.file.src_uri
if path in self.post_map:
context["posts"] = self.post_map[path]
if self.config.blog_toc:
self._populate_toc(page, context["posts"])
# Create pagination
pagination = paginate.Page(
self.post_pages,
page = list(self.post_map.keys()).index(path) + 1,
items_per_page = self.config.pagination_per_page,
url_maker = lambda n: utils.get_relative_url(
self.post_pager_pages[n - 1].url,
page.url
)
)
# Create pagination pager
context["pagination"] = lambda args: pagination.pager(
format = self.config.pagination_template,
show_if_single_page = False,
**args
)
# Provide post excerpts for archive
if self.config.archive:
if path in self.archive_post_map:
context["posts"] = self.archive_post_map[path]
if self.config.archive_toc:
self._populate_toc(page, context["posts"])
# Provide post excerpts for categories
if self.config.categories:
if path in self.category_post_map:
context["posts"] = self.category_post_map[path]
if self.config.categories_toc:
self._populate_toc(page, context["posts"])
# Determine whether we're running under dirty reload
def on_serve(self, server, *, config, builder):
self.is_dirtyreload = self.is_dirty
# -------------------------------------------------------------------------
# Generate and register files for archive
def _generate_files_for_archive(self, config, files):
for path, meta in self.post_meta_map.items():
file = files.get_file_from_path(path)
if self._is_draft(path):
continue
# Compute name from format string
date_format = self.config.archive_date_format
name = self._format_date(meta["date"], date_format, config)
# Compute path from format string
date_format = self.config.archive_url_date_format
path = self.config.archive_url_format.format(
date = self._format_date(meta["date"], date_format, config)
)
# Create file for archive if it doesn't exist
path = self._resolve(path + ".md")
if path not in self.archive_map:
self.archive_map[path] = []
# Generate and register file for archive
self._generate_file(path, f"# {name}")
self._register_file(path, config, files)
# Assign current post to archive
self.archive_map[path].append(file)
# Return generated archive files
return list(self.archive_map.keys())
# Generate and register files for categories
def _generate_files_for_categories(self, config, files):
allowed = set(self.config.categories_allowed)
for path, meta in self.post_meta_map.items():
file = files.get_file_from_path(path)
if self._is_draft(path):
continue
# Ensure category is in (non-empty) allow list
categories = set(meta.get("categories", []))
if allowed:
for name in categories - allowed:
log.error(
f"Blog post '{file.src_uri}' uses a category "
f"which is not in allow list: {name}"
)
sys.exit(1)
# Traverse all categories of the post
for name in categories:
path = self.config.categories_url_format.format(
slug = self.config.categories_slugify(
name, self.config.categories_slugify_separator
)
)
# Create file for category if it doesn't exist
path = self._resolve(path + ".md")
if path not in self.category_map:
self.category_map[path] = []
# Generate and register file for category
self._generate_file(path, f"# {name}")
self._register_file(path, config, files)
# Link category path to name
self.category_name_map[name] = path
# Assign current post to category
self.category_map[path].append(file)
# Sort categories alphabetically (ascending)
self.category_map = dict(sorted(self.category_map.items()))
# Return generated category files
return list(self.category_map.keys())
# -------------------------------------------------------------------------
# Check if a post is a draft
def _is_draft(self, path):
meta = self.post_meta_map[path]
if not self.config.draft:
# Check if post date is in the future
future = False
if self.config.draft_if_future_date:
future = meta["date"] > datetime.now()
# Check if post is marked as draft
return meta.get("draft", future)
# Post is not a draft
return False
# Generate a post excerpt relative to base
def _generate_excerpt(self, file, base, config, files):
page = file.page
# Generate temporary file and page for post excerpt
temp = self._register_file(file.src_uri, config)
excerpt = Page(page.title, temp, config)
# Check for separator, if post excerpt is required
separator = self.config.post_excerpt_separator
if self.config.post_excerpt == "required":
if separator not in page.markdown:
log.error(f"Blog post '{temp.src_uri}' has no excerpt.")
sys.exit(1)
# Ensure separator at the end to strip footnotes and patch h1-h5
markdown = "\n\n".join([page.markdown, separator])
markdown = re.sub(r"(^#{1,5})", "#\\1", markdown, flags = re.MULTILINE)
# Extract content and metadata from original post
excerpt.file.url = base.url
excerpt.markdown = markdown
excerpt.meta = page.meta
# Render post and revert page URL
excerpt.render(config, files)
excerpt.file.url = page.url
# Find all anchor links
expr = re.compile(
r"<a[^>]+href=['\"]?#[^>]+>",
re.IGNORECASE | re.MULTILINE
)
# Replace callback
first = True
def replace(match):
value = match.group()
# Handle anchor link
el = fragment_fromstring(value.encode("utf-8"))
if el.tag == "a":
nonlocal first
# Fix up each anchor link of the excerpt with a link to the
# anchor of the actual post, except for the first one that
# one needs to go to the top of the post. A better way might
# be a Markdown extension, but for now this should be fine.
url = utils.get_relative_url(excerpt.file.url, base.url)
if first:
el.set("href", url)
else:
el.set("href", url + el.get("href"))
# From now on reference anchors
first = False
# Replace link opening tag (without closing tag)
return tostring(el, encoding = "unicode")[:-4]
# Extract excerpt from post and replace anchor links
excerpt.content = expr.sub(
replace,
excerpt.content.split(separator)[0]
)
# Determine maximum number of authors and categories
max_authors = self.config.post_excerpt_max_authors
max_categories = self.config.post_excerpt_max_categories
# Obtain computed metadata from original post
excerpt.authors = page.authors[:max_authors]
excerpt.categories = page.categories[:max_categories]
# Return post excerpt
return excerpt
# Generate a file with the given template and content
def _generate_file(self, path, content):
content = f"---\nsearch:\n exclude: true\n---\n\n{content}"
utils.write_file(
bytes(content, "utf-8"),
os.path.join(self.temp_dir, path)
)
# Register a file
def _register_file(self, path, config, files = Files([])):
file = files.get_file_from_path(path)
if not file:
urls = config.use_directory_urls
file = File(path, self.temp_dir, config.site_dir, urls)
files.append(file)
# Mark file as generated, so other plugins don't think it's part
# of the file system. This is more or less a new quasi-standard
# for plugins that generate files which was introduced by the
# git-revision-date-localized-plugin - see https://bit.ly/3ZUmdBx
file.generated_by = "material/blog"
# Return file
return file
# Register and populate a page
def _register_page(self, file, config, files):
page = Page(None, file, config)
_populate_page(page, config, files)
return page
# Populate table of contents of given page
def _populate_toc(self, page, posts):
toc = page.toc.items[0]
for post in posts:
toc.children.append(post.toc.items[0])
# Remove anchors below the second level
post.toc.items[0].children = []
# Translate the given placeholder value
def _translate(self, config, value):
env = config.theme.get_env()
# Load language template and return translation for placeholder
language = "partials/language.html"
template = env.get_template(language, None, { "config": config })
return template.module.t(value)
# Resolve path relative to blog root
def _resolve(self, *args):
path = posixpath.join(self.config.blog_dir, *args)
return posixpath.normpath(path)
# Format date according to locale
def _format_date(self, date, format, config):
return format_date(
date,
format = format,
locale = config.theme["language"]
)
# -----------------------------------------------------------------------------
# Helper functions
# -----------------------------------------------------------------------------
# Search the given navigation section (from the configuration) recursively to
# find the section to host all generated pages (archive, categories, etc.)
def _host(nav, path):
# Search navigation dictionary
if isinstance(nav, dict):
for _, item in nav.items():
result = _host(item, path)
if result:
return result
# Search navigation list
elif isinstance(nav, list):
if path in nav:
return nav
# Search each list item
for item in nav:
if isinstance(item, dict) and path in item.values():
if path in item.values():
return nav
else:
result = _host(item, path)
if result:
return result
# Copied and adapted from MkDocs, because we need to return existing pages and
# support anchor names as subtitles, which is pretty fucking cool.
def _data_to_navigation(nav, config, files):
# Search navigation dictionary
if isinstance(nav, dict):
return [
_data_to_navigation((key, value), config, files)
if isinstance(value, str) else
Section(
title = key,
children = _data_to_navigation(value, config, files)
)
for key, value in nav.items()
]
# Search navigation list
elif isinstance(nav, list):
return [
_data_to_navigation(item, config, files)[0]
if isinstance(item, dict) and len(item) == 1 else
_data_to_navigation(item, config, files)
for item in nav
]
# Extract navigation title and path and split anchors
title, path = nav if isinstance(nav, tuple) else (None, nav)
path, _, anchor = path.partition("#")
# Try to retrieve existing file
file = files.get_file_from_path(path)
if not file:
return Link(title, path)
# Use resolved assets destination path
if not path.endswith(".md"):
return Link(title or os.path.basename(path), file.url)
# Generate temporary file as for post excerpts
else:
urls = config.use_directory_urls
link = File(path, config.docs_dir, config.site_dir, urls)
page = Page(title or file.page.title, link, config)
# Set destination file system path and URL from original file
link.dest_uri = file.dest_uri
link.abs_dest_path = file.abs_dest_path
link.url = file.url
# Retrieve name of anchor by misusing the search index
if anchor:
item = SearchIndex()._find_toc_by_id(file.page.toc, anchor)
# Set anchor name as subtitle
page.meta["subtitle"] = item.title
link.url += f"#{anchor}"
# Return navigation item
return page
# -----------------------------------------------------------------------------
# Data
# -----------------------------------------------------------------------------
# Set up logging
log = logging.getLogger("mkdocs.material.blog")