This is a CRITICAL bug fix release for everyone who is on the 6.4.x
series. Please upgrade to avoid potential data loss! The version has
been bumped to 6.5.0, please let everyone know to stay away from 6.5.x!
I am sorry for this.
See details in the Changelog and even more gory details in commit
message for commit 8fc7227189.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
This change looks harmless, but it fixes a severe bugfix, potentially
leading to data loss! It fixes the "on n new uploads, it will redownload
n-1, n-2, n-3,... messages during the next syncs" condition, and this is
what happens:
If there are more than one Mails to upload to a server, we do that by
repeatedly invoking folder.IMAP.savemessage(). If the server supports
the UIDPLUS extension we query the resulting UID by doing a:
imapobj._get_untagged_response('APPENDUID', True)
and that is exactly the problem. The "True" part causes the reply to
remain in the "response stack" of the imaplib2 library. When we do
the same call on a subsequent message and the connection is still on the
same folder, we will get the same UID response back (imaplib2 only looks
for the first matching response and returns that). The only time we
clear the response stack, is when the IMAP connection SELECTS a
different folder.
This means that when we upload 10 messages, the IMAP server gives us
always the same UID (that of the first one) back. And trying to write
out 10 different messages with the same UID will confuse OfflineIMAP.
This is the reason why we saw the ongoing UPLOADING/DOWNLOADING behavior
that people reported. And this is the reason why we saw the
inconsistency in the UID mapping in the IMAP<->IMAP case.
I urge everyone to upgrade ASAP. Sorry for that, I don't know why the
problem only became prevalent in the recent few releases as this code
has been there for quite a while.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Rather than always parsing the filename, we only need to do so if the flags
have actually changed, otherwise we can keep the filename.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Previously, assigning a new UID to a mapped IMAP or Maildir repository
was done by loading the "local" item, saving it under a new UID and
deleting the old one. This involved lots of disk activity for nothing
more than an effective file rename in Maildirs, and lots of network
usage in the MappedUID cases.
We do this on every upload from a local to a remote item, so that can
potentially be quite expensive. This patch lets backends that support it
(Maildir, MappedUID) efficiently rename the file rather than having to
read the mail content, write it out as a new file and delete the old
file. This speeds up uploads from Maildir and the MappedUID server.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Various functions (such as change_message_uid) will want to construct
maildir filenames, so factor out the code into a helper.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Create a helper function that retrieves the UID, folder MD5, and Flags from
a message filename.
We need these items when we simply want to rename (=new UID) a Maildir
message file later. The new function can give us these components.
Rework, so we cache the calculation of the folder's md5 value once, it
never changes and we call it a lot.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
If someone had a custom :2,a flag, adding a new flag would lead to the
invalid maildir filename ...a:2,... due to regex deficiencies not coping
with this. Fix this so we alway produce valid maildir names.
Note that custom flags are still problematic: as the syncing to the
remote IMAP server will fail, the next sync will assume that they have
been removed from the remote IMAP side and they will be removed from the
local Maildir then. We will need to think about how to handle this. At
least, with this patch we won't lose standard flags and won't produce
invalid maildir names.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
The regex for catching Maildir message flags was
self.infosep + '.*2,([A-Z]+)' (infosep being ':').
The .* is bogus, as there is nothing between the : and the 2, per
maildir name specification, so remove that unneeded piece.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
When a new remote folder was detected, we tried to create the folder
locally on the Maildir and called repository.forgetfolders() to force a
new scanning of the Maildir. However, that implementation used the
inherited base function that did nothing. We simply needed to implement
forgetfolders() to set self.folder=None, so we would force a new read in
of the updated local folder structure.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
1) Rename the unintuitive repository.syncfoldersto() to
sync_folder_structure()
2) We were checking if the local repository is readonly and then turning
off any folder creation. But as we can create folders on a remote
repository too, we need to be more fine grained here. Just don't create
a folder on the repository that is marked readonly=True.
This still does not do away with the error message that one currently
gets on missing local folders.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
The quiet UI should only output errors, and the final "Finished account
X in 2 seconds" clearly is none, so the message debug level needed to be
reduced to INFO to suppress it in the quiet ui.
Fixes https://github.com/spaetz/offlineimap/issues/6
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Merge in the new logging mechanism, and provide an --info feature that
will help with debugging. There is still some unstableness, so there
will be another release soon, but this should be no worse than 6.4.2 at
least...
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Previously, we would simply bail out in an ugly way, potentially leaving
temporary files around etc, or while writing status files. Hand SIGINT
and SIGTERM as an event to the Account class, and make that bail out
cleanly at predefined points. Stopping on ctrl-c can take a few seconds
(it will e.g. finish to transfer the ongoing message), but it will shut
down cleanly.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
As reported in https://github.com/spaetz/offlineimap/pull/2, we would
fail when files are empty because file.read() would throw attribute
errors.
Fix this by removing the superfluous read() check and additionally log
some warning message.
Reported-by: Ralf Schmitt
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
We need the list of folders and the folder delimiter, but it was not
always retrieved early enough. E.g. when doing IMAP<->IMAP sync and the
local IMAP being readonly, we would bunk out with a mysterious error
message become repository.getsel() would still return None.
This commit fixes this error.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
When sep='/' in a Maildir, we were doing a os.path.join(dirname,'') on
the top level maildir, which results in a "dirname/", so all our maildir
folder names had slashes appended. Which is pretty much wrong, so this
fixes it by only using os.path.join when we actually have something to
append.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
When sep='/' in a Maildir, we were doing a os.path.join(dirname,'') on
the top level maildir, which results in a "dirname/", so all our maildir
folder names had slashes appended. Which is pretty much wrong, so this
fixes it by only using os.path.join when we actually have something to
append.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
We need the list of folders and the folder delimiter, but it was not
always retrieved early enough. E.g. when doing IMAP<->IMAP sync and the
local IMAP being readonly, we would bunk out with a mysterious error
message become repository.getsel() would still return None.
This commit fixes this error.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
imapserver.getdelim() was not used at all, so remove this function. The
folder delimiter is available via the repository.getsep() call.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
The folder delimiter is only initialized after a call to
acquireconnection(), so we must never call this function too
early. Include an assert() to make sure we get notified when we do.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Rather than keeping a separate queue of all logged lines in memory, we
rely on the curses window scrolling functionality to scroll lines. On
resizing the terminal this means, we'll clear the screen and start
filling it afresh, but that should be acceptable.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
1) Rework the sleep abort request to set the skipsleep configuration
setting that the sleep() code checks.
2) Only output 15 rather than 50 debug messages on abort...
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
The remote|local|statusrepo is an anttribute of each SyncableAccount()
anyway, so we don't need to pass it in, we can simply get it from the
Account().
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Rename some variables, simplify the hotkeys treatment. Refresh/exit
signals still don't work as of yet, but will come.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Resizing a Blinkenlights terminal doesn't crash anymore, and actually
seems to be changing the size, with this patch.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
We were still referring to s.gettf() in sleeping(self, ...) causing each
attempt to sleep to crash. Fix this, and the CursesAccountFrame.sleeping()
method. I am sure, there is still wrong and broken but we are getting there.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
This function can IMHO lead to possible deadlocks when waiting for the
connectionlock. Do add a comment to that regard, this will need to audit.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
The huge UI rework patch removed some obscure logic and special handling of
thread exit messages. It turns out that this was in fact still needed as a
specific exit message of the SyncRunner thread signified the threatmonitor
to quit.
We will want a nicer machinery for this in the future I guess, but fix the
eternal hang on exit by reintroducing a special exit message for the
SyncRunner thread, and return from the infinite monitor loop if SyncRunner
finishes.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
To make sure, the lock gets released even if we raise an exception between
acquire() and release()
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
All ExitNotifyThreads and InstanceLimitThreads are setDaemon(True) in their
constructor, so there is no need to do that again in the code.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
