summaryrefslogtreecommitdiffstats
path: root/python/fatcat_tools/cleanups/files.py
diff options
context:
space:
mode:
authorBryan Newbold <bnewbold@robocracy.org>2019-10-08 15:56:53 -0700
committerBryan Newbold <bnewbold@robocracy.org>2019-10-08 15:56:53 -0700
commit76db7f4048116a23c82bdd70bb11dd004e347e8e (patch)
treeb273f321e4645121e2579de5d1478e7722bf629f /python/fatcat_tools/cleanups/files.py
parentb3bba513a843029459823ce9a74cce9947bba339 (diff)
downloadfatcat-76db7f4048116a23c82bdd70bb11dd004e347e8e.tar.gz
fatcat-76db7f4048116a23c82bdd70bb11dd004e347e8e.zip
new cleanup python tool/framework
Diffstat (limited to 'python/fatcat_tools/cleanups/files.py')
-rw-r--r--python/fatcat_tools/cleanups/files.py74
1 files changed, 74 insertions, 0 deletions
diff --git a/python/fatcat_tools/cleanups/files.py b/python/fatcat_tools/cleanups/files.py
new file mode 100644
index 00000000..c2733ba0
--- /dev/null
+++ b/python/fatcat_tools/cleanups/files.py
@@ -0,0 +1,74 @@
+
+from fatcat_openapi_client.rest import ApiException
+from fatcat_openapi_client.models import FileEntity
+from fatcat_tools.transforms import entity_to_dict, entity_from_json
+
+from .common import EntityCleaner
+
+
+class FileCleaner(EntityCleaner):
+ """
+ File fixups!
+ """
+
+ def __init__(self, api, **kwargs):
+
+ eg_desc = kwargs.pop('editgroup_description',
+ "Automated cleanup of file entities (eg, remove bad URLs)")
+ eg_extra = kwargs.pop('editgroup_extra', dict())
+ eg_extra['agent'] = eg_extra.get('agent', 'fatcat_tools.FileCleaner')
+ super().__init__(api,
+ entity_type=FileEntity,
+ editgroup_description=eg_desc,
+ editgroup_extra=eg_extra,
+ **kwargs)
+
+ def clean_entity(self, entity):
+ """
+ TODO: mimetype is bogus like (???) => clean mimetype
+ """
+
+ # URL has ://web.archive.org/web/None/ link => delete URL
+ entity.urls = [u for u in entity.urls if not '://web.archive.org/web/None/' in u.url]
+
+ # URL has ://archive.org/ link with rel=repository => rel=archive
+ for u in entity.urls:
+ if '://archive.org/' in u.url and u.rel == 'repository':
+ u.rel = 'archive'
+
+ # URL has short wayback date ("2017") and another url with that as prefix => delete URL
+ stub_wayback_urls = []
+ full_wayback_urls = []
+ for u in entity.urls:
+ if '://web.archive.org/web/' in u.url:
+ if len(u.url.split('/')[4]) <= 8:
+ stub_wayback_urls.append(u.url)
+ else:
+ full_wayback_urls.append('/'.join(u.url.split('/')[5:]))
+ for stub in stub_wayback_urls:
+ target = '/'.join(stub.split('/')[5:])
+ if target in full_wayback_urls:
+ entity.urls = [u for u in entity.urls if u.url != stub]
+
+ return entity
+
+ def try_update(self, entity):
+
+ try:
+ existing = self.api.get_file(entity.ident)
+ except ApiException as err:
+ if err.status != 404:
+ raise err
+ self.counts['skip-not-found'] += 1
+ return 0
+
+ if existing.state != 'active':
+ self.counts['skip-existing-inactive'] += 1
+ return 0
+ if existing.revision != entity.revision:
+ self.counts['skip-revision'] += 1
+ return 0
+
+ self.update_file(self.get_editgroup_id(), entity.ident, entity)
+ return 1
+