aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor/common.py
diff options
context:
space:
mode:
authordirkf <fieldhouse@gmx.net>2023-07-19 14:14:50 +0100
committerdirkf <fieldhouse@gmx.net>2023-07-19 22:14:50 +0100
commitb2ba24bb026904f3503db71f65d2b1627f08edf1 (patch)
treea81697902ace6dbdf90d0140f4a3067b4844e048 /youtube_dl/extractor/common.py
parenta190b559640ce1b5fe67e5a4843dc58328503f3c (diff)
[InfoExtractor] Add `_match_valid_url()` class method and refactor
* API compatible with yt-dlp * also support Sequence of patterns in _VALID_URL * one place to compile _VALID_URL * TODO: remove existing extractor shims
Diffstat (limited to 'youtube_dl/extractor/common.py')
-rw-r--r--youtube_dl/extractor/common.py51
1 files changed, 38 insertions, 13 deletions
diff --git a/youtube_dl/extractor/common.py b/youtube_dl/extractor/common.py
index 549781186..7f416d312 100644
--- a/youtube_dl/extractor/common.py
+++ b/youtube_dl/extractor/common.py
@@ -83,6 +83,7 @@ from ..utils import (
urljoin,
url_basename,
url_or_none,
+ variadic,
xpath_element,
xpath_text,
xpath_with_ns,
@@ -371,9 +372,22 @@ class InfoExtractor(object):
title, description etc.
- Subclasses of this one should re-define the _real_initialize() and
- _real_extract() methods and define a _VALID_URL regexp.
- Probably, they should also be added to the list of extractors.
+ A subclass of InfoExtractor must be defined to handle each specific site (or
+ several sites). Such a concrete subclass should be added to the list of
+ extractors. It should also:
+ * define its _VALID_URL attribute as a regexp, or a Sequence of alternative
+ regexps (but see below)
+ * re-define the _real_extract() method
+ * optionally re-define the _real_initialize() method.
+
+ An extractor subclass may also override suitable() if necessary, but the
+ function signature must be preserved and the function must import everything
+ it needs (except other extractors), so that lazy_extractors works correctly.
+ If the subclass's suitable() and _real_extract() functions avoid using
+ _VALID_URL, the subclass need not set that class attribute.
+
+ An abstract subclass of InfoExtractor may be used to simplify implementation
+ within an extractor module; it should not be added to the list of extractors.
_GEO_BYPASS attribute may be set to False in order to disable
geo restriction bypass mechanisms for a particular extractor.
@@ -409,21 +423,32 @@ class InfoExtractor(object):
self.set_downloader(downloader)
@classmethod
- def suitable(cls, url):
- """Receives a URL and returns True if suitable for this IE."""
-
+ def __match_valid_url(cls, url):
# This does not use has/getattr intentionally - we want to know whether
- # we have cached the regexp for *this* class, whereas getattr would also
- # match the superclass
+ # we have cached the regexp for cls, whereas getattr would also
+ # match its superclass
if '_VALID_URL_RE' not in cls.__dict__:
- cls._VALID_URL_RE = re.compile(cls._VALID_URL)
- return cls._VALID_URL_RE.match(url) is not None
+ # _VALID_URL can now be a list/tuple of patterns
+ cls._VALID_URL_RE = tuple(map(re.compile, variadic(cls._VALID_URL)))
+ # 20% faster than next(filter(None, (p.match(url) for p in cls._VALID_URL_RE)), None) in 2.7
+ for p in cls._VALID_URL_RE:
+ p = p.match(url)
+ if p:
+ return p
+
+ # The public alias can safely be overridden, as in some back-ports
+ _match_valid_url = __match_valid_url
+
+ @classmethod
+ def suitable(cls, url):
+ """Receives a URL and returns True if suitable for this IE."""
+ # This function must import everything it needs (except other extractors),
+ # so that lazy_extractors works correctly
+ return cls.__match_valid_url(url) is not None
@classmethod
def _match_id(cls, url):
- if '_VALID_URL_RE' not in cls.__dict__:
- cls._VALID_URL_RE = re.compile(cls._VALID_URL)
- m = cls._VALID_URL_RE.match(url)
+ m = cls.__match_valid_url(url)
assert m
return compat_str(m.group('id'))