duplicity-team team mailing list archive

[Kenneth Loafman <kenneth@xxxxxxxxxxx>]

It seems we are fighting an old Python bug still extant in Python 3.
Namely, sys.getfilesystemencoding() will return ascii if the LC_* variables
are not set (cron or other detached processes).  At one time in the early 3
series they defaulted to utf-8 if ascii was returned.  Then, as I
understand it, the purists won and ascii is returned now.  So, I think that
was a good enough idea, except we should allow an override.  I suggest we
allow an option if the FS is really something other than utf-8, but do
something like this in globals.py.

fse = sys.getfilesystemencoding()
globals.fsencoding = fse if fse != 'ascii' else 'utf-8'

Then allow it to be overridden in command line processing if needed.
Replace the two sys.getfilesystemencoding() with globals.fsencoding and we
should be 99% there.

...Ken





On Wed, Nov 15, 2017 at 9:40 AM, Kenneth Loafman <kenneth@xxxxxxxxxxx>
wrote:

> Google for 'tox getfilesystemencoding' and 'setup.py test
> getfilesystemencoding'.  You'll see a bunch of discussion.  It may be that
> we need to move from 'setup.py test' to something else.
>
> ...Ken
>
>
> On Tue, Nov 14, 2017 at 3:40 PM, Aaron <lists@xxxxxxxxxxxxxxxxxx> wrote:
>
>> Hello all,
>>
>> I am hoping for some help to iron out a small testing bug with:
>> https://code.launchpad.net/~aaron-whitehouse/duplicity/08-unicode
>> so that I can get the code committed. I believe that the code is working
>> correctly, but our test setup (tox, pexpect etc) is creating an environment
>> identified as ASCII rather than UTF-8 and that makes the tests fail.
>> *The branch aims to ease Python 2/3 compatibility*
>>
>> For context, this branch aims to ease the conversion of duplicity to be
>> Python 2/3 compatible. It looks to me as though the key stumbling block in
>> previous efforts has been the string unicode/bytes distinction in Python 3.
>> My plan with this branch was therefore to take manageable sections of
>> duplicity and convert the strings to Python 2 unicode/bytes strings, making
>> it much easier to then convert that code to Python 3 in the future, but in
>> a way that can be committed straight away to the existing code base.
>> *Using sys.getfilesystemencoding() misdetects 'ascii' in tests*
>>
>> As the branch currently stands, all tests pass. If, however (on my UTF-8
>> system) you change (util.py, line 66):
>>
>> return bytes_filename.decode("UTF-8", "ignore")
>>
>> to:
>>
>> return bytes_filename.decode(sys.getfilesystemencoding(), "ignore")
>>
>> then tests (mainly in testing.functional.test_selection.TestUnicode)
>> fail. Changing "ignore" in the above line to "strict" gives errors that
>> suggest an encoding error issue:
>>
>> UnicodeDecodeError: 'ascii' codec can't decode byte 0xd0 in position 25: ordinal not in range(128)
>>
>> This is the case even though sys.getfilesystemencoding() returns "UTF-8"
>> on my setup. Putting a print statement showing the result of
>> sys.getfilesystemencoding() shows this changing from "UTF-8" to
>> "ANSI_X3.4-1968" once the code is within the:
>>
>> child = pexpect.spawn(b'/bin/sh', [b'-c', cmdline.encode(sys.getfilesystemencoding(),
>>                                                          'replace')], timeout=None)
>>
>> *This looks to just be a problem with the test suite*
>>
>> The test suite prints a copy of the failing command, for example (from
>> testing.functional.test_selection.TestUnicode.test_unicode_
>> paths_non_globbing):
>>
>> ...command: "setsid" "-w" "duplicity" "full" "testfiles/select-unicode" "file://testfiles/output" "--volsize" "1" "--exclude" "testfiles/select-unicode/прыклад/пример/例/Παράδειγμα/उदाहरण.txt" "--exclude" "testfiles/select-unicode/прыклад/пример/例/Παράδειγμα/דוגמא.txt" "--exclude" "testfiles/select-unicode/прыклад/пример/例/მაგალითი/" "--include" "testfiles/select-unicode/прыклад/пример/例/" "--exclude" "testfiles/select-unicode/прыклад/пример/" "--include" "testfiles/select-unicode/прыклад/" "--include" "testfiles/select-unicode/օրինակ.txt" "--exclude" "testfiles/select-unicode/**" "-v0" "--no-print-statistics" "--allow-source-mismatch" "--archive-dir=testfiles/cache" < /dev/null
>>
>> If this (with PYTHONPATH added and the duplicity path adjusted, executed
>> from the "testing" folder with "testfiles.tar.gz" extracted) is run
>> directly in the commandline:
>>
>> $ PYTHONPATH=../ "../bin/duplicity" "full" "testfiles/select-unicode" "file://testfiles/output" "--volsize" "1" "--exclude" "testfiles/select-unicode/прыклад/пример/例/Παράδειγμα/उदाहरण.txt" "--exclude" "testfiles/select-unicode/прыклад/пример/例/Παράδειγμα/דוגמא.txt" "--exclude" "testfiles/select-unicode/прыклад/пример/例/მაგალითი/" "--include" "testfiles/select-unicode/прыклад/пример/例/" "--exclude" "testfiles/select-unicode/прыклад/пример/" "--include" "testfiles/select-unicode/прыклад/" "--include" "testfiles/select-unicode/օրինակ.txt" "--exclude" "testfiles/select-unicode/**" "-v0" "--no-print-statistics" "--allow-source-mismatch" "--archive-dir=testfiles/cache"
>>
>> Everything works correctly and restoring the files (using this branch)
>> and manually checking shows it worked correctly (even with "strict"). The
>> print statement also shows that the system encoding is "UTF-8" throughout.
>> Help requested
>>
>> Can anybody suggest what I can do to force the testing environment to be
>> UTF-8, or at least be detected as such by sys.getfilesystemencoding?
>> Alternatively, what is the least awful way to make the tests work enough to
>> get the (apparently working) code committed?
>> Many thanks,
>>
>> Aaron
>>
>> _______________________________________________
>> Mailing list: https://launchpad.net/~duplicity-team
>> Post to     : duplicity-team@xxxxxxxxxxxxxxxxxxx
>> Unsubscribe : https://launchpad.net/~duplicity-team
>> More help   : https://help.launchpad.net/ListHelp
>>
>>
>