When trying to load a torrents.state from version 1.3 users were
encountering the following error:
UnicodeDecodeError: 'ascii' codec can't decode byte
This was due to the way that Python 2 was pickling state with torrent
filenames that contained non-ascii characters and Python 3 was
unpickling the state using ascii encoding and failing. The fix is to
specify utf-8 encoding when loading torrents.state.
Some users have been reporting unhandled UnicodeDecodeErrors and the
traces show it occuring in the call to `peer.client`. Although unable to
replicate it seems prudent to put a try..except around the call to
ensure it does not break the UIs.
Refs: https://github.com/arvidn/libtorrent/issues/3858Closes: #3279
- Use sphinxcontrib.spelling with custom wordlist.
- Skip the checking of the modules documents as they raise
false-positives.
- Add a setup.py spellcheck_docs command.
- Fix spelling and other issues.
- Add a doc favicon.
Changes in libtorrent 1.1 mean that alerts are no longer allowed to be
accessed after the next call to pop_alerts.
> It is safe to call pop_alerts from multiple different threads, as
long as the alerts themselves are not accessed once another thread
calls pop_alerts. Doing this requires manual synchronization between
the popping threads.
The solution is to copy the alert attributes and pass that to the
handlers.
Refs: https://github.com/arvidn/libtorrent/issues/2779#3159
Some changes between lt 1.1 and 1.2 require updates to core code.
- Switch from proxy_type to proxy_type_t
- Replace hardcoded flag value with add_torrent_params_flags_t since
1.2 uses different flag values.
- add_torrent_params requires flags set instead of dict values.
Refs: #3255
When adding magnets that have been prefetched the tracker details were
lost. A result of returning only the lt.torrent_info.metadata which
does not contain full torrent details, such as trackers.
- Modified torrentmanager prefetch_metadata to return dict instead of
base64 encoded bencoded metadata dict...
- Used a namedtuple to ease identifying tuple contents.
- Updated tests to reflect changes with mock trackers added to
test_torrent.file.torrent.
- Refactor TorrentInfo to accept dict instead of bytes and add
a class method to accept metadata dict with lists of trackers.
- Rename class arg from metainfo to torrent_file, matching
lt.torrent_info.
- Rename metadata property to correct name; metainfo.
- Simplify class variable naming with _filedata and _metainfo for
torrent file contents encoded and decoded respectively.
- Update GTK Add torrent dialog to pass trackers to TorrentInfo.
Disable the warn_msg in daemon_entry since the argparse option are all
marked for translation and warnings are being output. The original
intent of this warn_msg was to identify anything in core code that was
incorrectly marked for translation.
- Move baseargparser out of deluge/ui since it is also used by the
Daemon and could cause packaging issues if UI code is not available.
- Renamed baseargparser to argparserbase to follow existing Deluge
naming.
- Renamed get_version to distinguish from deluge.common.get_version.
- Translation code is usable by more than just the UIs so also move it
to Deluge namespace and re-use i18n directory and make it a package.
- Renamed setup_translations to singular as it felt more correct.
- Renamed set_dummy_trans to be more descriptive.
Closes: #3081
- Fix new flake8 warnings from latest version.
Note: The `addSlash` variable was orphaned with no reference in
Twisted or Deluge code so removed.
- Update pre-commit config
- New pinned versions.
- Fix prettier output.
- Use new flake8 hook config and add naming plugin.
Removed previous workaround to ensure sync of file priorities with
libtorrent. This did not work when loading torrents as the status is
called before setting the file priorites and resets them to default.
Removed the call to set_file_priorities when writing the torrent file
to disk as it resets the options to default so although the torrent
file priorities do not change, the priorities for UIs is incorrect.
- Fixed the core not correctly settings the current file_priority
settings and added a test.
- Fixed the console not setting file priorities.
- Change the label for not downloading of a file to 'Skip'.
If the torrent.state was corrupted then loading would create a new
state with no backup to examine.
The solution is to use the archive function to save a copy of the
torrent.state.
Added a message argument to archive_files so that the error message
with a reason for archiving can be included in the tarball.
- Fix applying the setting to libtorrent, passing the value without
modification so it decide when to enable it.
- Enable super_seeding option when adding torrents to core.
- Update UIs with option in tabs and add dialogs.
- Found an issue while fixing `get_name` where `handle.rename_file`
would raise a UnicodeDecodeError with non-ascii on Python 2. The
fix is to catch this and pass unicode string to method instead.
- Add a test `test_rename_unicode` to verify no errors are generated.
- Updated test to use core.session instead of creating another one.
The recent change to torrent.get_name does not handle non-ascii paths
on Python 2.
- Add a decode_bytes to resolve the issue.
- Add tests.
- Refactor to reduce nesting.
- Add an About window to see version details like GTKUI.
- The author and license text were left out as unnecessary.
- Added a daemon get_version method since daemon version was not
available through the json-api.
- Fix LookupResource to ensure path exists when rendering.
The following error was encountered in GTK3 which is a result of trying
to cast a very large ETA value to C int and raising an Overlflow error.
<type 'exceptions.OverflowError'>: 3072227291 not in range -2147483648 to 2147483647
The solution is to limit the ETA to 1 year and represent any values over
that as -1 which the UIs can display as infinity.
The new get_name method needs to function as it did in 1.3-stable so
when renaming a torrent file or top-level folder update torrent name
to reflect that. If UI supports renaming the torrent then the options
value with be used instead.
The move to using auto-formatter makes it easier to read, submit and
speeds up development time. https://github.com/ambv/black/
Although I would prefer 79 chars, the default line length of 88 chars
used by black suffices. The flake8 line length remains at 120 chars
since black does not touch comments or docstrings and this will require
another round of fixes.
The only black setting that is not standard is the use of double-quotes
for strings so disabled any formatting of these. Note however that
flake8 will still flag usage of double-quotes. I may change my mind on
double vs single quotes but for now leave them.
A new pyproject.toml file has been created for black configuration.
Simplify the methods by initialising the session_status dict using
libtorrent session_stats_metrics and rate keys.
Instead of first looking for deprecated keys use exception then lookup.
Added a few more tests.
If a magnet is added to new Deluge state, then deleted, it will reappear
on restart.
The problem results from torrents requiring both state and torrent file
but magnet only rely on the state file and the save_state code not
saving if the torrent list is empty. So torrents won't be loaded as
their torrent files have been deleted but magnets details remain in
state file and are loaded again on restart.
The fix is to always save the state file even if the state is empty.
- Disable the magnet from being auto_managed so it starts immediately.
- Reduce the default timeout to 30secs.
- Use the generic tempfile dir.
- Move callback method to be class method
A magnet that prefetched the metadata is added as a normal torrent but
with the filename set to the magnet URI. So we need to assign the URI to
magnet arg and set filename to None before creating Torrent object.
Updated the is_magnet function to prevent AttributeError with None type.
The file priorities were not updating correctly in the UI and it was
found that in lt 1.1 file priorities are now updated asynchonously so we
cannot get the values immediately. So only update the options
file_priorities if they are empty.
This applies the following for both WebUI and Daemon:
1. Raised minimal TLS version to TLSv1.2
2. Added specific cipher suite list
3. Added support for ECDSA auth keys
4. Added support for ECDHE key exchange algorithm
We disabled the ability to perform TLS/SSL renegotiation and therefore
will prevent the clients from renegotiating, which can be exploit for
DoS attacks.
New security tests now will be skipped when running `pydef` and `trial`
testenvs. To run the test, use the testenv `security` or add the environment
variable `SECURITY_TESTS` before running the tests. Also should only run when
adding to the commit message the string `SECURITY_TEST`.
If the torrent_id argument received in the pause or resume methods is not a string, the methods execute with the un-parsed input and then with parsed input on a second call. A key error exception is thrown from the first call, and the second call succeeds.
Floor division will return a float if a float is provided so ensure int
when dividing by the stop_ratio. All other status values from libtorrent
are ints.
Added tests.
The following error occured on Windows when switching to using HTTPS
url with Twisted Agent:
```
<class 'twisted.web._newclient.ResponseNeverReceived'>: [<twisted.python.failure.Failure OpenSSL.SSL.Error: [('SSL routines', 'tls_process_server_certificate', 'certificate verify failed')]>
```
The fix is to install certifi and provide the path to the trust store as
env var for OpenSSL to pick up.
Also includes a simplication of the core test_listen_port code.
The usual minor fixes for unicode/bytes for library calls.
The minimum Twisted version is now 16 for Python 3 support so remove old
code and start replacing deprecated methods.
Raised the minimum TLS version to 1.2 for the web server.
* In Py3 base64.encodestring is deprecated so rather than use the
Py3 only encodebytes instead use b64encode. The other advantage is
that with issue a consistent TypeError is raised that we can catch.
