1. There is one database per folder and sqlite requires to serialize the
writings. Instead of locking at LocalStatusSQLiteFolder object level, introduce
a new DatabaseFileLock object which is shared across threads. This fixes the
concurrent writes issues that some users might experience by duplications or
flags restored to the previous state.
2. Close the database only when we are sure no other threads will use the
connection on a *per-file* basis. Previous fix 677afb8d8f is wrong
because the same lock is shared for all the databases.
Github-fix: https://github.com/OfflineIMAP/offlineimap/issues/350
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
It's required to have more than one connection to the database for the
maxconnections configuration option to work with threads. However,
connection.close() is closing all the connections. Only close the connection
when no more thread need it.
Backported-from: 856b74407bd7f634cae5a8c2d9b84e13d14c12d2
Github-fix: https://github.com/OfflineIMAP/offlineimap/issues/350
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
Backported from 8e995a69bfa003ab822b55731429d84b3bc5626f.
We currently close the database as soon as possible while we handle the status
backend but this is still too early because autorefresh induces looping on this
code block.
Instead of delaying the closing outside of the loop, it's easier to delay the
opening as late as possible (inside the loop). The downside is that the database
is opened/closed more than once when autorefresh is enabled. The good news is
that this make the code much easier.
Fixes regression introduces by 6fb5700.
Reported-by: Łukasz Żarnowiecki <dolohow@outlook.com>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
We usually mutate some exceptions to OfflineImapError() and it is
a whole lot better if such exception will show up with the original
traceback, so all valid occurrences of such mutations were transformed
to the 3-tuple form of "raise". Had also added coding guidelines
document where this re-raise strategy is documented.
Signed-off-by: Eygene Ryabinkin <rea@codelabs.ru>
Create initializer function that puts default values to all fields
of message list item. Fix all code that directly assigns some hash
to the elements of messagelist: for direct assignments only initializer
is now permitted, all other modification are done in-place.
Signed-off-by: Eygene Ryabinkin <rea@codelabs.ru>
If when we request a LocalStatus folder, the folder has to be created,
we look whether the other backend has data, and if it does we migrate
it to the new backend.
The old backend data is left untouched, so that if you change back say
from sqlite to plaintext, the older data is still there. That should
not lead to data loss, only a slower sync while the status folder gets
updated.
Signed-off-by: Eygene Ryabinkin <rea@codelabs.ru>
* Do not inherit LocalStatusSQLiteFolder class from the plaintext
one.
* Use some functions already in BaseFolder in both, plaintext and
sqlite classes.
* Add a saveall method. The idea is that saveall dumps the entire
messagelist to disk, while save only commits the uncommited
changes. Right now, save is noop for sqlite, and equivalent to
saveall for plaintext, but it enables to be more clever on when we
commit to disk in the future.
Signed-off-by: Eygene Ryabinkin <rea@codelabs.ru>
When synclabels config flag is set to "yes" for the GMail repo,
offlineimap fetches the message labels along with the messages, and
embeds them into the body under the header X-Keywords (or whatever
'labelsheader' was set to), as a comma separated list.
It also adds an extra pass to savemessageto, that performs label
synchronization on existing messages from GMail to local, the same way
it is done with flags.
We also introduce GmailMaildir repository that adds functionality to
change message labels. It keeps track of messages modification time,
so one can quickly detect when the labels may have changed.
Signed-off-by: Eygene Ryabinkin <rea@codelabs.ru>
Make external API of class/module to be smaller, explicitely mark
all internal functions. Also annotate methods that are implemented
as the part of the parent class interface.
Signed-off-by: Eygene Ryabinkin <rea@codelabs.ru>
They are redundant in all pruned cases and sometimes even create some
problems, e.g., when one tries to jump through paragraphs in vi.
Signed-off-by: Eygene Ryabinkin <rea@codelabs.ru>
Prevent savemessage(), and savemessageflags() to occur in dryrun mode in
all backends. Still need to protect against deletemessage().
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
'set' is builtin since python2.6, so remove the imports. Also 'ssl' exists
since 2.6 and has everything we need, so no need for conditional import
tests here anymore.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
When deleting many (eg 2000) mails using the SQLITE backend, this takes
a long time durig which OfflineImap can not be aborted via
CTRL-C. Thinking it had frozen permanently, I killed it hard, leaving a
corrupted db journal (which leads to awkwards complaints by OLI on
subsequent starts!). That shows that delete performance is critical and
needs improvement.
We were iterating through the list of messages to delete and deleted
them one-by-one execute()'ing a new SQL Query for each message. This
patch improves the situation by allowing us to use executemany(), which
is -despite still being one SQL query per message- much faster. This is
because rather than performing a commit() after each mail, we now do
only one commit() after all mails have been deleted.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
They have the Repository() which contains the root, so no need to pass
it in as an extra parameter. Rename repository.LocalStatus()'s
self.directory to self.root for consistency with other backends.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
It is possible to get the config parameter from the Repository() which is
set in BaseFolder, so we set self.config there and remove the various
methods and 'config' parameters that are superfluous.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
We passed in the accountname to all derivatives of BaseFolder, such as
IMAPFolder(...,repository,...,accountname), although it is perfectly
possible to get the accountname from the Repository(). So remove this
unneeded parameter. Each backend had to define getaccountname() (although
the function is hardly used and most accessed .accountname directly).
On the other hand BaseFolder was using getaccountname but it never defined
the function. So make the sane thing, remove all definitions from backends
and define accountname() once in Basefolder. It was made a property and not
just a (public) attribute, so it will show up in our developer
documentation as public API.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
We were using self.getfolderbasename(self.name) but the API is simply
getfolderbasename(). Fix this glitch.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
as that method doesn't exist on sets. Rather call the inbuilt
sorted(flags). This fixes Exceptions being thrown when using the sqlite
backend.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
As this is essentially what it is, a set of values. This allows as
to do set arithmetics to see, e.g. the intersection of 2 flag sets
rather than clunkily having to do:
for flag in newflags:
if flag not in oldflags:
oldflags.append(flag)
Also some more code documenting.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
Empty foldernames (as they could be created through nametrans) were
failing as the uidvalidity and status files names as determined by
folder/Base.py:getfolderbasename() lead to invalid file names ''.
Fix this by handling empty file names and translating them to '.' which
leads to the special file name 'dot'. (this special value existed before
and was not invented by this patch)
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
The syntax was not right, and deleting messages from the LocalStatus
failed. (We passed in the full list of uids and we need to pass in one
uid at a time (as a tuple). Deleting messages works now.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
Test if sqlite is multithreading-safe and bail out if not. sqlite
versions since at least 2008 are.
But, as it still causes errors when 2
threads try to write to the same connection simultanously (We get a
"cannot start transaction within a transaction" error), we protect
writes with a per class, ie per-connection lock. Factor out the retrying
to write when the database is locked.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
doautosave was a useless variable before (it was *always* 1). So we
remove the self.dofsync variable and store in doautosave whether we
should fsync as often as possible (which really hurts performance).
The sqlite backend could (at one point) use the doautosave variable to
determine if it should autocommit after each modification.
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
Based on patches by Stewart Smith, updated by Rob Browning.
plus:
- Inherit LocalStatusSQLFolder from LocalStatusFolder
This lets us remove all functions that are available via our ancestors
classes and are not needed.
- Don't fail if pysql import fails. Fail rather at runtime when needed.
- When creating the db file, create a metadata table which contains the
format version info, so we can upgrade nicely to other formats.
- Create an upgrade_db() function which allows us to upgrade from any
previous file format to the current one (including plain text)
Signed-off-by: Sebastian Spaeth <Sebastian@SSpaeth.de>
Signed-off-by: Nicolas Sebrecht <nicolas.s-dev@laposte.net>
