Issues for on non-english OS error text

[totaam]

Issue migrated from trac ticket # 1856

component: client | priority: trivial | resolution: wontfix

2018-05-28 14:01:22: stdedos created the issue

---

2018-05-28 15:39:27,405 Xpra gtk2 client version 2.3-19246 64-bit
2018-05-28 15:39:27,409  running on Microsoft Windows 10
2018-05-28 15:39:28,913 Warning: failed to import opencv:
<Unicode console stdout>.write: UnicodeDecodeError('utf8', '2018-05-28 15:39:28,913  DLL load failed: ??p??a ???t??a p??et??\xb5as?a? \xb5?a? \xdf?\xdf???????? d??a\xb5???? s??des?? (DLL) ap?t??e.\n', 64, 65, 'invalid start byte')
Traceback (most recent call last):
  File "C:/msys64/mingw64/lib/python2.7/logging/__init__.py", line 891, in emit
UnicodeDecodeError: 'utf8' codec can't decode byte 0xb5 in position 64: invalid start byte
Logged from file log.py, line 132
2018-05-28 15:39:28,913  webcam forwarding is disabled

I have tried various ways to go around that, but I cannot seem to find something special.

I think at some point I manage to "decode" the text, but I cannot seem to remember the used codepage. (I will update if I manage to)

Text is in Greek (suggested codepage for chcp is 1253)

Recommended chcp for UTF-8 is 65001 (although internet says it's broken anyway)

[totaam]

2018-05-28 14:01:27: stdedos uploaded file test-orig.log (1.0 KiB)

[totaam]

This exception comes from opencv, not xpra itself.
I have just checked and xpra handles unicode exception messages just fine.

So I will probably close this ticket as wontfix unless someone else fixes it.
We'll have to wait for #1231 to replace opencv with something better. ie something that doesn't break with such regularity - see r17571 and many other fixes like it..

[totaam]

2018-05-29 20:58:17: stdedos commented

---

I didn't manage to decode the file. Is it possible that the output is already mangled?

I'll add some 2c, but I am fine with your opinion regardless.
I have some example text in the mentioned encoding (see attached file).

It could be possible to "fix this" by adding cmd /V /C "set LANG=C && gpg -k --check-trustdb" (check on the attached file again).
Maybe this workaround is specific to gpg itself however?

[totaam]

2018-05-29 20:58:41: stdedos uploaded file examples.log (0.4 KiB)

[totaam]

2018-05-29 21:02:16: stdedos commented

---

(forgot to mention: in Notepad2 [http://www.flos-freeware.ch/notepad2.html], working encoding is OEM (737) / Greek (DOS-737)

[totaam]

I didn't manage to decode the file. Is it possible that the output is already mangled?

Yes, it's very likely.

It could be possible to "fix this" by adding cmd /V /C "set LANG=C && gpg -k --check-trustdb"

Using LANG=C is cheating, you're using the default locale.

[totaam]

2018-05-30 07:22:34: stdedos commented

---

Replying to [comment:5 Antoine Martin]:

It could be possible to "fix this" by adding cmd /V /C "set LANG=C && gpg -k --check-trustdb"
Using LANG=C is cheating, you're using the default locale.

Probably I am cheating. However, if that one-line workaround "saves" you some time until you switch from OpenCV once and for all, I am not sure if it would be too harmful.

Obviously, your call :D

[totaam]

Please try a build r19535 or later from [https://xpra.org/beta/windows].

[totaam]

Unfortunately, trick does not work :/
Apologies for the work.

[totaam]

2018-05-31 15:58:00: stdedos uploaded file xpra-cv-fail.log (0.7 KiB)

[totaam]

r19543 will set LC_ALL=C rather than LANG=C - updated builds to come later.
You can already check if that is going to work by setting it before running xpra.

If that still doesn't help, then I will close this ticket as "wontfix".

[totaam]

2018-05-31 16:09:17: stdedos commented

---

Oups! ... Maybe should've thought that earlier :/

However; this doesn't work. I literally did:

set LC_ALL=C
set LANG=C
xpra attach [...]

Does not parse it.

Since you went through the trouble to push a new commit, I'll test the build when it comes. However, not too optimistic about it :/