During startup we can get above 1000 alerts at each pop even with only
30 torrents in the queue. This is because libtorrent will post
piece_finished_alert and file_completed_alert for each torrent. These
alerts push out of the way the ones we care about.
The alert queue will be grown to max only if needed. So we don't use
more memory. It will greatly depend on how many torrents a user has in
their session.
When getting fastresume_rejected_alert we need to act as fast as
possible in pausing it, otherwise there's a chance it will begin
downloading and writing to disk before we pause it.
`QIODevice::read(qint64 maxSize)` will allocate full `maxSize` of memory no matter
what the real file size was, this caused users to experience out-of-memory
exception on 32-bit qbt.
Also handle the OOM execption if it still fails.
Closes#9064, #9075, #9130, #9239, #9246, #9279.
"Active torrents" is a somewhat unintuitive concept as a basis for
preventing sleep, as torrents can become active or inactive on the
network at any time. This brings some predictability to the inhibit
sleep option, and will inhibit sleep as long as there are unpaused
downloads or uploads, regardless of network activity.
Closes#1696, #4592, #4655, #7019, #7159, #7452
Updating file priorities is an async operation in libtorrent, when we
just updated it and immediately query it, we might get the
old/wrong values, so we rely on `updatedFilePrio` in this case.
Don't post "torrent resumed" event when torrent starts in "resumed"
state.
Fix confusing names. Now "resumed torrent" means "unpaused torrent"
only. When we load previously added torrent it is called "restored
torrent".
When a file error happens libtorrent spews a ton of `file_error_alert`
which result in log floods and notification balloon floods. The later
might render the program inaccessible because the constant
notifications prevent the user from interacting with the tray icon.
Closes#8934
Now it applies the setting to all files in torrent, no matter whether
the file is previewable or not.
Torrent client shouldn't be smart by deciding which file this
option will be applied to, i.e. it should just follow the user's decision.
When having big torrents, the rechecking might take a while to
complete. Users need some sort of feedback that progress is being made.
Also, I changed the float to qreal since that's the return type.
This is another indication to the user that something is happening
behind the scenes.
Uses the same icon/color as "Checking" status.
Torrents in the `Moving` state are considered as "Active". This should
prevent accidental program exit from the user and inhibit system sleep.
Libtorrent can recheck only unpaused torrents. We get around this by
unpausing the torrent, issuing the recheck and pausing again after we
get alerted by libtorrent that the recheck has finished. This alert is
asyncronous. There is a small time frame where the program might start
downloading and writing data to the file before we pause it. This can
lead to data corruption if the file on disk is totally different that
the one expected by the torrent AND the file on disk is a valid file on
its own. OR in case the user points the new torrent to the wrong
directory by mistake.
To get around this the torrent is placed in upload_mode and out of
automanagement.
Using iostream usually adds a lot of other operators (<<, endl), whereas
*printf takes only 1 function call.
Also use qUtf8Printable whenever possible.
Stops temporary containers being created needlessly due to API misuse.
For example, it’s common for developers to assume QHash::values() and
QHash::keys() are free and abuse them, failing to realize their
implementation internally actually iterates the whole container, allocates
memory, and fills a new QList.
Added a removeIf generic algorithm, similar to std ones. We can't use std
algorithms with Qt dictionaries because Qt iterators have different
behavior from the std ones.
Found using clazy.
`torrent_info` constructor has default limits that can't be changed via
parameters, so we handle the loading process manually and explicitly
specifiy the limits to `bdecode()`.
The token_limit is also changed to 10000000.
Now it is defined as:
CacheStatus.readRatio = (blocks read from cache) / (blocks read from disk + blocks read from cache)
The 2 variables in denominator are counted separately and the formula before this change doesn't really make sense
Add percentage sign to "Read cache hits" stats
Also remove redundant header include
Normalize Web API method names.
Allow to use alternative Web UI.
Switch Web API version to standard form (i.e. "2.0").
Improve Web UI translation code.
Retranslate changed files.
Add Web API for RSS subsystem.
