| Age | Commit message (Collapse) | Author |
|---|
|
Although this draft-saving feature seems to be somehow brittle, since it
breaks from time to time, the causes are different and subtle on each
case.
This time the bug is related to having the notify_just_mdoc (or
fast_notifies as I'm tempted to call it in the future for more clarity)
set to True by default in the APPENDS.
Fast notifies break the "save draft" functionality because what
Thunderbird does is an Append of the newest message, followed by
deletion of the old message and a SEARCH by Message-ID. For this we need
the headers-doc to be already inserted in the store.
Since the fast-notify makes a *big* difference in terms of insertion
times for serialized appends, I've opted for using a heuristic for
detecting if it's the case of a Draft message, using a mozilla-specific
header. If this is found, we set the notify_just_mdoc unconditionally to
False. We'll deal with other MUAs on its due time.
Releases: 0.4.0
|
|
the bug consist on a fetch-while-pending-inserts hanging. the pending
insert dict was not being cleaned up because the lookup for the
Message-Id *is* case-sensitive (in the headers dict).
by using a temporary all-keys-lowercase dict the lookup can be performed
right, and the fetch returns successfully.
at this point there's still a pending bug with Drafts, and it is that
the new version is inserted but the MUA (TB) doesn't hide the older
version (although a Delete flag is added).
Resolves: #7189, #7190
Releases: 0.4.0
|
|
- Add errback handling to catch properly errors that were not allowing
the complete insertion of the parts for a given message. Fixes blank
attachments and moving of messages to different folders.
- Force overwritting of mdoc when it is a copy. This was avoiding a
message to be copied back to a folder from where it already had been
copied to another (since the mdoc was already existing there, with the
same doc_id, which was forbidding the creation of the new one).
This case also needs special care in the indexer, since we have to
delete the old hash entry first.
Closes: #7178, #7158
|
|
|
|
Closes: #7169.
|
|
|
|
because we love to test with mutt.
|
|
|
|
before, we were taking the length of a string, signalling an incorrect
number to the ui.
currently this event is not being used, just only logged. in the future
the ui could probably might want to make use of this info to keep record
of a separate counter (how many mails received in the last sync).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
using the new soledad plugin capablity, mail hooks to the post-sync
event by subscribing to the Meta-Doc type of documents. In this way, we
can create the uid tables and the uid entries needed to keep local
indexes for mail that has been processed in another instance.
however, this won't prevent a conflict if a given mail is received and
processed in two different instances. that is a problem that we still
have to deal with.
Resolves: #6996
Releases: 0.4.0
|
|
- Related: #6359
|
|
Current twisted implementation correctly parses partial fetches using
substrings by use of angle brackets (see section 6.4.5 of imap rfc), but
no use is made of the requested substring in the spew_body method.
this commit minimally implements conformance to the substring request,
although further boundary checks should be made (ie, checking whether
the starting octet is beyond the end of the text).
Resolves: #6841
Releases: 0.4.0
|
|
|
|
There's a workaround for "slow" APPENDS to an inbox, and it is that we
have a flag to allow returning early when JUST the mdoc (the
meta-document) has been written. However, this was givin a problem when
doing a FETCH right after an APPEND (with notify_just_mdoc=True) has
been done.
This commit fixes it by making the FETCH command first check if there's
an ongoing pending write, and queueing itself right after the write
queue has been completed.
This fixes the testFullAppend regression.
Releases: 0.4.0
|
|
This also fixes a bug in which INBOX wasn't being given a creation
timestamp, and therefore always being identified with the same
UIDVALIDITY = 1, which could be confusing MUAs since this value should
be unique, and it's relied on to uniquely identifying a given message.
Releases: 0.4.0
|
|
just in case
Releases: 0.4.0
|
|
Until we implement sequences, this avoids breaking with certain MUAs
like mutt.
Releases: 0.4.0
|
|
This fix stores as multi-line headers that are repeated, and that were
being discarded when storing them in a regular dict.
It had been removed during the last refactor.
I also store headers now as a case-insensitive dict, which solves other
problems with the implementation of the twisted imap.
Releases: 0.4.0
|
|
The MIME size is the size of the body, w/o counting the headers.
Releases: 0.4.0
|
|
By removing this call to lower(), we avoid a bug in which the
BODYSTRUCTURE response returns a boundary all in lower case.
Should send patch upstream to twisted.
Related: #6773
|
|
Thunderbird (as of 37.0b1) will display a blank body (with no
attachments) if some conditions are met:
* disk synchronization is disabled
* mime_part_on_demand = true
* msg size is bigger than the parts_on_demand threshold (30000 by
default).
Comparing the logs with a well behaved imap server (dovecot, on this
case), it's easy to see that twisted implementation is lacking an extra
line separator at the end of each group of headers that is rendered in
response to each of the `BODY.PEEK[X.MIME]` command that the
mime_parts_on_demand will issue after getting the BODYSTRUCTURE.
This change patches the spew_body command on the body server.
We still would have to see if this is a bad behaviour in the thunderbird
side. The most similar bug I've found is:
https://bugzilla.mozilla.org/show_bug.cgi?id=149771
Which apparently was happening with exchange server.
We should send the patch to upstream twisted as well.
Note that this fix is not enough: the following commit, about fixing the
case of the boundary passed in the BODYSTRUCTURE response is also needed
to fix the bug (since a bad parsing happens all the same).
Resolves: #6773, #5010
Documentation: #6773
Releases: 0.4.0
|
|
Catch null doc_id so that we don't interrupt server
This bug needs further investigation
Related: #6769 ?
|
|
it is a weakref dictionary so that the collections can be garbage
collected when out of scope.
Releases: 0.4.0
|
|
this test was failing randomly because we were returning the deferred
before all the documents were saved into soledad store.
changed also the delete_msg deferred chaining for better readability.
Releases: 0.9.0
|
|
We're raising the exception now, not a Failure.
Releases: 0.9.0
|
|
because the commits can be prettier this way :)
|
|
while updating the tests, I found that IMAPMessageCollection was not
actually being used: all the work is done in IMAPMailbox, using directly
the MessageCollection instance. So, this extra level of abstraction was
finally not used.
Releases: 0.9.0
|
|
|
|
it is interpreted as "tags" otherwise.
|
|
|
|
|
|
|
|
bug: delete_msg was still passing the mbox_name instead of the mbox_uuid
as it should.
|
|
* fix get_next_uid test
* remove duplication of maybe_first_query_item, since get_last_uid also
do it now.
|
|
* Set the internal date from within the incoming mail service.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
