Issue1693

Specifically, names with non-ascii characters in them. Whether the module you're trying to import exists or not is irrelevant.

CPython 2.5.5:
>> sys.path.append(u'/home/alex/t/töö')
>>> import ttt
>>> 

Jython 2.5.2rc2:
>>> sys.path.append(u'/home/alex/t/töö')
>>> import ttt
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/alex/libs/jython2.5.2rc2/Lib/encodings/__init__.py", line 31, in <module>
    import codecs, types
UnicodeEncodeError: 'ascii' codec can't encode characters in position 14-15: ordinal not in range(128)

The unit test in the attached file test_sys2_jy.py fails,
regardless if the first line
  # coding=latin2
is present or not.

The changes in 1693-patch.txt solve the problem.

The fix is to encode a unicode string with latin-1 instead of ascii only (in __str__()).
javatests and regrtests all pass.

pjenvey: could you please review this? - thanks!

I'm sorry to say that this patch doesn't cut it, not by far.
Two reasons: first, using latin-1 encoding in PyUnicode breaks CPython compatibility (UnicodeError should be thrown when u'åäö' is converted to str); second, you'd still get a UnicodeError when adding a path element with, say, chinese characters. Why are sys.path elements being converted to bytestrings anyway?

Alex is right. Our import system is converting sys.path items to java Strings via item.__str__().toString(). CPython in this case converts unicode to strings by encoding them via the filesystem encoding.

We don't support a filesystem encoding on Jython (at this point anyway). Instead we've just been 'passing thru' unicode when it's requested (e.g. os.listdir).

That is technically broken (you could end up with a plain str where ord(somestr[0]) > 255) but I think we'll continue getting away with this strategy until 2.6. This is one of the few leftover str/unicode weirdness bits carried over from 2.2 where unicode and str were pretty much the same object

something like this..

applied that patch and Oti's test in r7182