locale.getdefaultlocale() fails on OS X

[jheinen]

For any reason locale.getdefaultlocale() fails on OS X (returns None), which subsequently produces errors and makes the matplotlib module unusable. Checking for a None result and then setting the default encoding to "utf-8" seems to help.

[WeatherGod]

Don't do equality when testing against None. Do if encoding is None:.

[WeatherGod]

@mdehoon, does this fallback make sense to you? I don't know about system encodings on Macs.

[jenshnielsen]

On my mac locale.getdefaultlocale() returns ('en_GB', 'UTF-8') so UTF-9 should be fine but there might be other OSs where this fails?

[mdehoon]

No idea, sorry.

[jheinen]

According to the Python documentation, the result of local.getdefaultlocale() can be None.
Excerpt from https://docs.python.org/2/library/locale.html:

Except for the code 'C', the language code corresponds to RFC 1766. language code and encoding may be None if their values cannot be determined.

So, matplotlib should check the result and fallback to a default encoding. This is a pragmatic solution. and certainly better than changing the locale or expecting an environment setting in the user's shell (LANGUAGE, LC_ALL, LANG, ...).

[tacaswell]

@jheinen Thanks, you short cut my research on this which was going through https://bugs.python.org/issue18378 !

I am 👍 on merging this, but as the bug was reported specifically against OSX, one of the OSX-based devs should merge this.

[efiring]

I am not an OSX expert by any means, but if no one else comments or merges it, I will do so within 24 hours. I don't see how it could hurt anything.

[jkseppan]

On an old Mac I have that is running OS X 10.8, locale.getdefaultlocale() returns (None, None). TextEdit defaults to UTF-8 on that Mac. Some discussion threads I find on the Internet suggest that this default was new with 10.7, and 10.6 might have used single-byte character sets by default.

I think Microsoft Office 2008 still used Mac Roman by default, so there's at least a theoretical risk if you edited your matplotlibrc in Word 2008 and used non-ASCII characters for some values.

But in any case, if passing encoding=None is problematic for UTF-8 files, my guess is that that affects a larger number of users than the inverse problem of passing encoding='UTF-8' when the file is actually Mac Roman or ISO 8859-1. I think it would be nice to give a meaningful error message to the user, though.

[jkseppan]

Plenty of software is going to fail strangely if your locale doesn't match your files. I guess utf-8 is a reasonable default in 2015, but I also think it's the user's or sysadmin's responsibility to fix the configuration.

[jkseppan]

I have a suggestion for a slightly better error message at jheinen#1.

[matthew-brett]

On a couple of 10.6 Macs I have:

Mac 1: (None, None) for Pythons 2.5, 2.5, ('en_US', 'UTF-8') for Python 3.4.1
Mac 2: (None, None) for Pythons 2.6 through 3.4.

[jkseppan]

I think this is useful as is, so I'll merge it, and make a separate PR about the error message.