diff options
author | Nikolaus Schulz <microschulz@web.de> | 2006-11-01 05:27:30 +0000 |
---|---|---|
committer | Nikolaus Schulz <microschulz@web.de> | 2006-11-01 05:27:30 +0000 |
commit | bea6ef63907ea86af68b907837abf92029a6a9a9 (patch) | |
tree | 07709c086ee45c98bd6c989f9fc24204be14a021 | |
parent | 86f1fec50df0ee39abf7b277f09eb353631df413 (diff) | |
download | archivemail-bea6ef63907ea86af68b907837abf92029a6a9a9.tar.gz archivemail-bea6ef63907ea86af68b907837abf92029a6a9a9.tar.bz2 archivemail-bea6ef63907ea86af68b907837abf92029a6a9a9.zip |
Optimized IMAP code: just invert the message list from the SEARCH response
instead of querying the server another time with an inverted search filter.
Gather RFC822.SIZE in one big run.
-rwxr-xr-x | archivemail.py | 58 |
1 files changed, 23 insertions, 35 deletions
diff --git a/archivemail.py b/archivemail.py index 7770e7a..f51141d 100755 --- a/archivemail.py +++ b/archivemail.py @@ -1018,7 +1018,7 @@ def is_older_than_days(time_message, max_days): return 1 return 0 -def build_imap_filter(invert = False): +def build_imap_filter(): """Return an imap filter string""" imap_filter = [] @@ -1041,13 +1041,7 @@ def build_imap_filter(invert = False): if options.filter_append: imap_filter.append(options.filter_append) - if not invert: - return '(' + string.join(imap_filter, ' ') + ')' - - imap_filter = map(lambda x: 'NOT ' + x, imap_filter) - if len(imap_filter) == 1: - return '(' + imap_filter[0] + ')' - return reduce(lambda x,y: '(OR ' + x + ' ' + y + ')', imap_filter) + return '(' + string.join(imap_filter, ' ') + ')' ############### mailbox operations ############### @@ -1303,9 +1297,7 @@ def _archive_imap(mailbox_name, final_archive_name): stats = Stats(mailbox_name, final_archive_name) imap_str = mailbox_name[mailbox_name.find('://') + 3:] imap_filter = build_imap_filter() - inverse_imap_filter = build_imap_filter(invert=True) vprint("imap filter: '%s'" % imap_filter) - vprint("inverse imap filter: '%s'" % inverse_imap_filter) try: imap_username, imap_str = imap_str.split('@', 1) imap_server, imap_folder = imap_str.split('/', 1) @@ -1332,44 +1324,40 @@ def _archive_imap(mailbox_name, final_archive_name): result, response = imap_srv.login(imap_username, imap_password) if result != 'OK': unexpected_error("authentication failure") vprint("logged in to server as %s" % imap_username) + result, response = imap_srv.select(imap_folder) if result != 'OK': unexpected_error("cannot select imap folder") - # response is e.g. ['1016'] for 1016 messages in folder vprint("selected imap folder %s" % imap_folder) - vprint("folder has %s message(s)" % response[0]) + # response is e.g. ['1016'] for 1016 messages in folder + total_msg_count = int(response[0]) + vprint("folder has %d message(s)" % total_msg_count) - result, response = imap_srv.search(None, inverse_imap_filter) + # IIUIC the message sequence numbers are stable for the whole session, since + # we just send SEARCH, FETCH and STORE commands, which should prevent the + # server from sending untagged EXPUNGE responses -- see RFC 3501 (IMAP4rev1) + # 7.4.1 and RFC 2180 (Multi-Accessed Mailbox Practice). + # Worst thing should be that we bail out FETCHing a message that has been + # deleted. + + result, response = imap_srv.search(None, imap_filter) if result != 'OK': unexpected_error("imap search failed") # response is a list with a single item, listing message ids # like ['1 2 3 1016'] message_list = response[0].split() - vprint("%d messages are not matching filter" % len(message_list)) + vprint("%d messages are matching filter" % len(message_list)) - max_fetch = 100 - for i in range(0, len(message_list), max_fetch): - result, response = imap_srv.fetch(string.join(message_list[i:i+max_fetch], ','), - '(RFC822.SIZE)') - if result != 'OK': unexpected_error("Failed to fetch message size") + # First, gather data for the statistics. + if total_msg_count > 0: + result, response = imap_srv.fetch('1:*', '(RFC822.SIZE)') + if result != 'OK': unexpected_error("Failed to fetch message sizes") # response is a list with entries like '1016 (RFC822.SIZE 3118)', # where the first number is the message id, the second is the size. for x in response: - msg_size = int(x.split()[2][:-1]) - stats.another_message(msg_size) - - result, response = imap_srv.search(None, imap_filter) - if result != 'OK': unexpected_error("imap search failed") - message_list = response[0].split() - vprint("%d messages are matching filter" % len(message_list)) - - for i in range(0, len(message_list), max_fetch): - result, response = imap_srv.fetch(string.join(message_list[i:i+max_fetch], ','), - '(RFC822.SIZE)') - if result != 'OK': unexpected_error("Failed to fetch message size") - for x in response: - # for the parsing magic see above - msg_size = int(x.split()[2][:-1]) + msn, blurb, msg_size = x.split() + msg_size = int(msg_size.rstrip(')')) stats.another_message(msg_size) - stats.another_archived(msg_size) + if msn in message_list: + stats.another_archived(msg_size) if not options.dry_run: if not options.delete_old_mail: |