The synchonous add torrent method was replaced with async but this
break backward compatibility with 3rd party plugins and clients.
Added a new add_torrent_file_async method for adding single torrent.
Torrent manager has a new add_async method and split up code to prevent
duplication.
Update any use of add_torrent_file to add_torrent_file_async. Future
refactoring could use add_torrent_files instead.
* The verfication of timestamps of files on disc against those in resume data
can be buggy and sends torrents to error state for no apparent reason. It has
also been removed in latest version of libtorrent so disabling this check.
- Preparation work for using six or future module for Py2/3 compat. The
code will be written in Python 3 with Python 2 fallbacks.
- Added some Py3 imports with Py2 fallbacks to make it easier to remove
Py2 code in future.
- Replace xrange with range (sort out import as top of files in future).
- Workaround Py2to3 basestring issue with inline if in instances. This means
every usage of basestring is more considered.
- Replace iteritems and itervalues for items and values. There might be a
performance penalty on Py2 so might need to revisit this change.
- Python 3 renames `unicode` type to `str` and introduces `bytes` type.
- Python 2.7 has `bytes` but is only an alias for `str` so restricted
to comparisons but helps keep compatibility.
- To test for unicode string on Py2 and Py3 uses the "''.__class__" type.
- Remove usage of utf8encode and just encode, problems with bytes being passed
in code will be picked up faster.
- Where possible refactor out isinstance for try..except duck-typing.
* Twisted methods require byte-string arguments.
* The headers str conversion in httpdownloader _download_file was
incorrent and left the dict key still as unicode so replaced with
a dict comprehension (py2.7 requirement).
* Added `from __future__ import unicode_literals` to every file so
now all strings in code are forced to be unicode strings unless
marked as byte string `b'str'` or encoded to byte string `'str'.encode('utf-8')`.
This is a large change but we have been working towards the goal of unicode
strings passed in the code so decoding external input and encoding
output as byte strings (where applicable).
Note that in Python 2 the `str` type still refers to byte strings.
* Replaced the use of `str` for `basestring` in isinstance comparison as
this was the original intention but breaks code when encoutering unicode strings.
* Marked byte strings in gtkui as the conversion to utf8 is not always handled, mostly
related to gobject signal names.
* A rather disruptive change but for a few reasons such as easier to read,
easier type, keep consistent and javascript code uses single quotes.
* There are a few exceptions for the automated process:
* Any double quotes in comments
* Triple double quotes for docstrings
* Strings containing single quotes are left e.g. "they're"
* To deal with merge conflicts from feature branches it is best to follow
these steps for each commit:
* Create a patch: `git format-patch -1 <sha1>`
* Edit the patch and replace double quotes with single except those in
comments or strings containing an unescaped apostrophe.
* Check the patch `git apply --check <patchfile>` and fix any remaining
issues if it outputs an error.
* Apply the patch `git am < <patchfile>`
To make the code more Python 3 compatible, I've made a few changes to how we handle keys() or iterkeys() calls on dictionaries. All functionality should remain the same.
* Remove the use of .keys() or .iterkeys() when iterating through a dictionary.
* Remove the use of .keys() when checking if key exists in dictionary.
* Replace dict.keys() with list(dict) to obtain a list of dictionary keys. In Python 3 dict.keys() returns a dict_keys object, not a list.
* libtorrent >= 1.1 dropped support for GeoIP so this adds support
again using MaxMind GeoIP Legacy Python Extension API.
For reference it is known by the following package names:
* Maxmind: geoip-api-python
* Linux: python-geoip
* PyPi: GeoIP
* Ensure all file descriptors are closed. Using the with statement ensures
closure.
* The main problem was with thousands of unclosed file desciptors from
tracker_icons mkstemp.
* Use a prefix 'deluge_ticon.' to identify created tracker_icon tmp files.
This should fix problems with errors occuring when failing to
enable plugins. Errors in plugin handling are handled better
and properly logged.
WebUI plugin in particular had issues when being enabled and disabled
multiple times because it was trying to create DelugeWeb component
each time it was enabled. If deluge-web is already listening on
the same port, enabling the WebUI plugin will fail, and the checkbox
will not be checked.
There are still some issues when enabling/disabling plugins by
clicking fast multiple times on the checkbox.
Selected Warning messages disabled in pylintrc:
* unused-argument: Quite a large and disruptive change if enabled.
* broad-except: Most required in-depth investigation to determine type.
* fixme: Not important
* protected-access: Complicated to fix
* import-error: Too many false-positives
* unidiomatic-typecheck: Should be fixed in the next round of checks.
* unused-variable: Again large and disruptive changes.
* global-statement: Most usage is required.
* attribute-defined-outside-init: Should be fixed in next round of checks.
* arguments-differ: Possible false-positives, needs revisited.
* no-init, non-parent-init-called, super-init-not-called: False-positives?
* signature-differs: False-positives?
* The paused state of torrents is now correctly stored on shutdown if the session is paused.
* Resume session refreshes all the torrents' state. This fixes only torrents that changed state being
updated so queued torrents would be incorrectly displayed as paused.
Removing multiple torrents with remove_torrents will be faster
than calling remove_torrent for each torrent, as the state file
will be written only once, after all the torrents are deleted.
