If running on the Japanese version of Windows XP or Vista, NVDA should displays the alert of OS version requirements

[nishimotz]

Link to issue number:

Closes #7586

Summary of the issue:

If running on the Japanese version of Windows XP or Vista, NVDA raises error as follows.

See the logfile 'C:\Users\***\AppData\Local\Temp\***.tmp\app\nvda_noUIAccess.exe.log' for details

Traceback (most recent call last):
  File "nvda.pyw", line 77, in <module>
UnicodeDecodeError: 'ascii' codec can't decode byte 0x8e in position 0: ordinal not in range(128)

It makes difficult for supporters to solve the problem of the users.

Description of how this pull request fixes the issue:

• if ctypes.FormatError() returns mbcs string, it should be decoded.
• the code is expected to work with both Python 2.7 and Python 3.7.

Testing performed:

• this workaround is already done for NVDA Japanese version 2018.3.1jp.

Known issues with pull request:

• similar workaround may be necessary for other parts of NVDA, so we should avoid the duplication.

Change log entry:

Section: Bug fixes

If running on the Japanese version of Windows XP or Vista, NVDA now displays the alert of OS version requirements as expected.

[nishimotz]

The appveyor build nvda_snapshot_pr8771-16089,8e12d46c.exe is tested on Windows Vista Japanese and worked as expected.

[LeonarddeR]

Personally, I'd rather have this addressed at the winUser.MessageBox level. You can add:

if isinstance(text, bytes)
    text = text.decode("mbcs")

Also, you could do the same to caption.

Than in NVDA.pyw, you can revert your change and change the MessageBox call as follows:

winUser.MessageBox(0, ctypes.FormatError(winUser.ERROR_OLD_WIN_VERSION), None, winUser.MB_ICONERROR)

[nishimotz]

@LeonarddeR Thank you for the review.
nvda_snapshot_pr8771-16090,2938b4c1.exe is tested and worked as well.

[LeonarddeR]

The use of isinstance instead of hasattr in my previous comment was intentional, see below.

[LeonarddeR]

Please use isinstance(text, bytes) instead. Otherwise, this will lead to unnecessary decoding on python 2 unicode strings.

[LeonarddeR]

You can safely use hasattr on a None, so the if caption check is not required. Furthermore, also avoid using hasattr here and use an instance check instead.

[LeonarddeR]

Actually, at least on python 3, I'm pretty sure that you must ommit the if caption check. If you don't and you provide an empty byte string, ctypes might throw an error, telling you that you can't throw an empty byte string into a function that expects a wide character string.

[nishimotz]

nvda_snapshot_pr8771-16091,73deebfe.exe is tested as well.

[LeonarddeR]

Thanks for addressing the review comments. Looks all good now.

[kvark128]

Strangely, this problem was described a year ago, but then it was suggested to close it:
#7586