Include the necessary ifdefs for DAV functionality #2
Commits on Feb 5, 2014
-
[ticket 303341] message: test 0 length fieldnames
it works just fine in the old message API but we may as well test for it.
-
-
-
-
-
-
-
-
sequence: return NULL is a seqset has no length
"" as a seqset really doesn't make sense anywhere, and it broke QRESYNC's VANISHED (EARLIER) response.
-
-
-
message_parse_csection: handle MESSAGE/RFC822 with multipart children
Re-build the structure in the same shape that it went in during message_write_section()
-
Revert "message_parse_csection: handle MESSAGE/RFC822 with multipart …
…children" This reverts commit f23947db227e6215a52697ad00a485a9aea83d76. Obviously still buggy despite passing tests, joy - will try again later
-
Revert "message: log IOERROR if failing to parse body part"
This reverts commit 6cf39343f3c3d08384c62ebf532ef994b7f14252. Just shut up the noise for now - but let's not forget it's broken!
-
-
-
message: explicit multipart check is better
Not only does it handle RFC822 inside an RFC822, but it's actually reflecting the same codepath that wrote the section in the first place.
-
xapian: writelock activefile on write to avoid retry errors
This should not be necessary in some future release of xapian, so we can remove it again then
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
cyr_dbtool: default to non-transactional
Safe is better than fast!
-
-
-
mboxevent: code tidy for _notify
I don't _think_ it changes anything, other than using a foreach loop and making the event list a bit more sensible about bunching flags.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
xapian: filter and reindex to new DB support
Including review from the maintainer. Three modes now. Regular mode is pure compact. Also -F for "filter", which will copy messages to a brand new database (non-compacted) using the cyrus.index unexpunged records, and -X for "reindex", which will actually parse the raw messages and index them again
-
-
-
-
-
-
-
-
cache: don't cache a small set of unused headers
This should shrink the sync size down to manageable!
-
-
-
-
-
-
-
-
-
-
-
-
-
-
imap: add command 'XKILLMY', usage like this:
TAG XKILLMY xconvmultisort Always returns 'OK', regardless of what was killed
-
-
-
-
-
-
-
-
-
-
it's invasively coupling, and not needed
-
flags: more clarity about fields in system_flags
Fixes some bugs in sync copyback in particular, but also a race condition in index.c which will matter later
-
mailbox: pass record to all filename lookups
This will matter later when we want to move files depending on flags in the record
-
index: track cache_offset in map
It can change if the message gets archived underneath us. The alternative is to read that system-flag from the disk, but then you get into the horrible land of write races, and partial reads, which locking protects us from. Better to just spend the extra 4 bytes (which rounds up to 8 - but we can get them ALL back if we're willing to play bitmap games with the empty space in system_flags)
-
This involves massive changes to caching, using mappedfile. It involves new config options. There's a bugzilla bug, which should be mentioned in here. For FastMail, we intend to hand the new messages out on SSD, and then shunt all the old ones off onto slow storage. It cuts the fsync count to the slow disks by a LOT.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
mailbox: recreate cache records if missing
(would fix today's incidents)
-
-
mailbox: create cache on save if required as well...
(and label for repack, because things are weird)
-
-
-
cyr_expire: kinda funky "per-user" mode
Should be ported to a real per-user thing at some point
-
-
-
-
sync_server: send file from archive if not in spool
NOTE: we don't check the index, we just blat one or the other - in THEORY it's slightly racy... because we don't open the file first...
-
-
-
-
-
search_xapian: don't try to scan mailbox tombstones (TEMPORARY)
This should be fixed by parsing and checking the entry before calling the callback
-
takes a single atom, i.e. TAG XMEID MYID And sets that ID for mboxevents
-
-
-
-
-
-
-
-
-
id: quirk os=iOS os-version=7...
Turn all searches into FUZZY searches to make BODY less crazy expensive
-
-
-
-
-
-
-
-
-
-
-
imap: create-special-use should create as the mailbox owner
not the authenticated (possibly admin) user
-
-
-
-
-
-
-
-
search: don't crash on modseq search:
e.g. 6 UID SEARCH RETURN (ALL COUNT) MODSEQ 4336 Which was crashing because there was no uint64_cmp function.
-
-
-
-
-
-
-
-
-
-
-
-
-
sync: don't suppress mailbox events just because user event exists
This should fix user rename issues (hopefully)
-
-
-
-
-
-
-
-
-
-
-
-
twoskip: keep a separate location tracker for 'foreach'
Avoids a ton of strcmp work
-
-
-
-
twoskip: don't be confused by chained deletes
(they can't exist now, but maybe one day, and it doesn't cost much to check)
-
-
-
-
-
-
-
-
-
-
twoskip: late-resolve offsets in readahead
The callback might have refreshed the mmap without changing the size (i.e. a rollback)
-
Revert "twoskip: late-resolve offsets in readahead"
This reverts commit 105aa22.
-
Revert "twoskip: switch back to squeeze compatible syntax"
This reverts commit 9159406.
-
Revert "twoskip: use skipdelete in consistency/recovery stages"
This reverts commit 63c132d.
-
Revert "stitch: don't write past what we read"
This reverts commit 43cf8dc.
-
Revert "twoskip: don't be confused by chained deletes"
This reverts commit 8f2d8f9.
-
Revert "twoskip: don't keep re-reading base and size pointers"
This reverts commit be8f332.
-
Revert "twoskip: readahead in foreach to reduce lockyness"
This reverts commit bbd8516.
-
Revert "twoskip: keep a separate location tracker for 'foreach'"
This reverts commit e6f3efd.
-
Revert "cyrusdb: only CRC check if integrity checks turned on"
This reverts commit 84f1fb3.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Commits on Feb 10, 2014
Commits on Feb 12, 2014
Commits on Feb 18, 2014
-
mailbox: don't lock conversations again during cleanup
It's not necessary, since nothing in cleanup uses it, and it's a lock inversion. This is one of a bunch of options here (including keeping it locked throughout the cleanup) - but it's the least invasive, and it's perfectly correct.
-
-
-