aboutsummaryrefslogtreecommitdiffstats
path: root/python/scripts
diff options
context:
space:
mode:
Diffstat (limited to 'python/scripts')
-rwxr-xr-xpython/scripts/archiveorg_fileset.py138
1 files changed, 138 insertions, 0 deletions
diff --git a/python/scripts/archiveorg_fileset.py b/python/scripts/archiveorg_fileset.py
new file mode 100755
index 0000000..0e507eb
--- /dev/null
+++ b/python/scripts/archiveorg_fileset.py
@@ -0,0 +1,138 @@
+#!/usr/bin/env python3
+"""
+Helper script to
+
+Takes either two args (release ident and archive.org item), or a stream of
+tab-separated such pairs on stdin.
+
+TODO:
+- should this check the item type?
+"""
+
+import sys
+import json
+from typing import Any
+
+import internetarchive
+
+
+FORMAT_TO_MIMETYPE = {
+ 'BZIP': 'application/x-bzip',
+ 'BZIP2': 'application/x-bzip2',
+ 'ZIP': 'application/zip',
+ 'GZIP': 'application/gzip',
+ 'RAR': 'application/vnd.rar',
+ 'TAR': 'application/x-tar',
+ '7z': 'application/x-7z-compressed',
+
+ 'HTML': 'text/html',
+ 'Text': 'text/plain',
+ 'PDF': 'application/pdf',
+
+ 'CSV': 'text/csv',
+ 'XML': 'application/xml',
+ 'JSON': 'application/json',
+
+ #'application/msword (.doc)', # .doc
+ #'application/vnd.openxmlformats-officedocument.wordprocessingml.document', # .docx
+ #'application/vnd.ms-excel', # .xls
+ #'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet', # .xlsx
+
+ 'MP3': 'audio/mpeg', # .mp3
+
+ 'MP4': 'video/mp4', # .mp4
+ 'MPEG': 'video/mpeg', # .mpeg
+
+ 'JPEG': 'image/jpeg',
+ 'GIF': 'image/gif',
+ 'PNG': 'image/png',
+ 'TIFF': 'image/tiff',
+
+ 'Unknown': None,
+}
+
+def want_file(f: dict, item_name: str) -> bool:
+ """
+ Filters IA API files
+ """
+ if f.source != 'original':
+ return False
+ for suffix in [
+ '_meta.sqlite',
+ '_archive.torrent',
+ '_itemimage.jpg',
+ '_meta.xml',
+ '_thumb.png',
+ '_files.xml',
+ ]:
+ if f.name == item_name + suffix or f.name == item_name.lower() + suffix:
+ return False
+ if f.name.startswith('_'):
+ return False
+ if item_name.startswith('academictorrents_'):
+ for suffix in ['_academictorrents.torrent', '_academictorrents_torrent.txt', '.bib']:
+ if f.name == item_name + suffix:
+ return False
+ return True
+
+def parse_file(f: dict) -> dict:
+ """
+ Takes an IA API file and turns it in to a fatcat fileset manifest file
+ """
+ assert f.name and f.sha1 and f.md5
+ assert f.name is not None
+ mf = {
+ 'path': f.name,
+ 'size': int(f.size),
+ 'sha1': f.sha1,
+ 'md5': f.md5,
+ }
+ # TODO: will disable this hard check eventually and replace with:
+ #mimetype = FORMAT_TO_MIMETYPE.get(f.format)
+ mimetype = FORMAT_TO_MIMETYPE[f.format]
+ if mimetype:
+ mf['extra'] = dict(mimetype=mimetype)
+ return mf
+
+def item_to_fileset(item_name: str, release_id: str, session: internetarchive.ArchiveSession):
+ print(f"processing item={item_name} release_id={release_id}", file=sys.stderr)
+ if release_id.startswith('release_'):
+ release_id = release_id[9:]
+ assert len(release_id) == 26
+ item = session.get_item(item_name)
+ assert item.metadata['mediatype'] not in ['collection', 'web']
+ item_files = item.get_files(on_the_fly=False)
+ manifest = [parse_file(f) for f in item_files if want_file(f, item_name)]
+ fileset = {
+ 'manifest': manifest,
+ 'urls': [
+ {
+ 'rel': 'archive',
+ 'url': f'https://archive.org/download/{item_name}/',
+ },
+ ],
+ 'release_ids': [release_id],
+ #extra={},
+ }
+ print(json.dumps(fileset))
+ return fileset
+
+def main():
+ session = internetarchive.get_session()
+ if len(sys.argv) == 3:
+ item_name = sys.argv[1]
+ release_id = sys.argv[2]
+ item_to_fileset(item_name, release_id=release_id, session=session)
+ else:
+ for line in sys.stdin:
+ line = line.strip()
+ if not line:
+ continue
+ fields = line.split('\t')
+ assert len(fields) == 2
+ item_name = fields[0]
+ release_id = fields[1]
+ item_to_fileset(item_name, release_id=release_id, session=session)
+
+if __name__ == '__main__':
+ main()