Expected behaviour:

When unset, the locale in use is `C` (as stated in python documentation) and `locale.getlocale()` returns  `(None, None)` on Linux with python2.7 or on Windows with python2.7 and python 3.6 (at least):


$ python2
Python 2.7.15 (default, May  1 2018, 20:16:04) 
[GCC 7.3.1 20180406] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import locale
>>> locale.getlocale()
(None, None)
>>> 


Issue:

But when using python3.4+ on Linux, instead of `(None, None)`, `locale.getlocale()` returns the same value as `locale.getdefaultlocale()`:


$ python
Python 3.6.3 (default, Oct 24 2017, 14:48:20) 
[GCC 7.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import locale
>>> locale.getlocale()
('fr_FR', 'UTF-8')
>>> locale.localeconv()
{'int_curr_symbol': '', 'currency_symbol': '', 'mon_decimal_point': '', 'mon_thousands_sep': '', 'mon_grouping': [], 'positive_sign': '', 'negative_sign': '', 'int_frac_digits': 127, 'frac_digits': 127, 'p_cs_precedes': 127, 'p_sep_by_space': 127, 'n_cs_precedes': 127, 'n_sep_by_space': 127, 'p_sign_posn': 127, 'n_sign_posn': 127, 'decimal_point': '.', 'thousands_sep': '', 'grouping': []}
>>> locale.str(2.5)
'2.5'


Though the locale actually in use is still `C` (as shown above by the output of `locale.localeconv()` and confirmed by the result of `locale.str(2.5)`, which shows a dot as decimal point and not a comma (as expected with `fr_FR.UTF-8`)).

I could observe this confusing behaviour on Linux with python3.4, 3.5, 3.6 and 3.7 (rc1). (Also on FreeBSD with python3.6.1).

A problematic consequence of this behaviour is that it becomes impossible to detect whether the locale has already been set by the user, or not.

I could not find any other similar issue and hope this is not a duplicate.

Can you say on which Linux platform/release you see this behavior and with which Python 3.6.3, i.e. from the platform distributor or built yourself?  If I understand your concern correctly, I cannot reproduce that behavior on a current Debian test system using either the Debian-supplied 3.6.6rc1 or with a  3.6.3 built from source.  With either LANG unset or set to C (and with no LC* env vars set), I see:

$ unset LC_ALL LC_CTYPE LANG LANGUAGE
$ ./python
Python 3.6.3 (tags/v3.6.3:2c5fed86e0, Jun 22 2018, 16:08:11)
[GCC 7.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import locale
>>> locale.getlocale()
(None, None)
>>> locale.getdefaultlocale()
(None, None)

Note that, as documented, the locale.getdefaultlocale() checks several env vars 'LC_ALL', 'LC_CTYPE', 'LANG' and 'LANGUAGE'.  Are you certain that all of those env vars are unset when you run this test?

https://docs.python.org/3.6/library/locale.html#locale.getdefaultlocale

Sorry, I did not realize that using the word "unset" was completely misleading: I only meant "before any use of locale.setlocale() in python". So I'll rephrase this all, and add details about the python versions and platforms in this message.

So, first, I do not unset the environment variables from the shell before running python.

The only steps required to reproduce this behaviour are: open a terminal and run python3:

Python 3.6.5 (default, May 11 2018, 04:00:52) 
[GCC 8.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import locale
>>> locale.getlocale()
('fr_FR', 'UTF-8')  # Wrong: the C locale is actually in use, so (None, None) is expected


Explanation: when python starts, it runs using the C locale, on any platform (Windows, Linux, BSD), any python version (2, 3...), until locale.setlocale() is used to set another locale. This is expected (the doc says so in the getdefaultlocale() paragraph that you mentioned) and can be confirmed by the outputs of locale.localeconv() and locale.str().

So, before any use of locale.setlocale(), locale.getlocale() should return (None, None) (as this value matches the C locale).

This is the case on Windows, python2 and 3, and on Linux and FreeBSD python2.

But on Linux and FreeBSD, python>=3.4 (could not test 3.0<=python<=3.3), locale.getlocale() returns the value deduced from the environment variables instead, like locale.getdefaultlocale() already does, e.g. ('fr_FR', 'UTF-8').

All python versions I tested are from the platform distributors (3.7 only is compiled, but it's an automatic build from an AUR). Here is a more detailed list of the python versions and Linux and BSD platforms where I could observe this behaviour:

- Python 3.4.8, 3.5.5, 3.6.5 and 3.7.0rc1 on an up to date Manjaro (with "LTS" kernel): Linux 4.14.48-2-MANJARO #1 SMP PREEMPT Fri Jun 8 20:41:40 UTC 2018 x86_64 GNU/Linux

- Python 3.6.5 on Xubuntu 18.04 (as virtual box guest) Linux 4.15.0-23-generic #25-Ubuntu SMP Wed May 23 18:02:16 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux

- Python 3.4.6 on openSUSE Leap 42.3 (as virtual box guest) Linux 4.4.76-1-default #1 SMP Fri Jul 14 08:48:13 UTC 2017 (9a2885c) x86_64 x86_64 x86_64 GNU/Linux

- Python 3.4.8 and 3.6.1 on FreeBSD 10.4-RELEASE-p8 FreeBSD 10.4-RELEASE-p8 #0: Tue Apr  3 18:40:50 UTC 2018     root@amd64-builder.daemonology.net:/usr/obj/usr/src/sys/GENERIC  amd64

Problem of this behaviour on Linux and FreeBSD python>=3.4 is first, of course, that it's not consistent throughout all platforms, and second, that it makes it impossible for a python library to guess, from locale.getlocale() if the user (a python app) has set the locale or not (and is actually still using the C locale). (It is still possible to rely on locale.localeconv() to get correct elements).

Hope this message made things clear now :-)

Thanks for the more detailed explanation.  I think you are right that the behavior does not match the documentation but which is to be preferred does not necessarily have an easy answer.  Also, this whole area has been undergoing revision, for example, with new features in 3.7.  Nick and/or Victor, can you address Nicolas's query?

This statement is no longer correct: "when python starts, it runs using the C locale, on any platform (Windows, Linux, BSD), any python version (2, 3...), until locale.setlocale() is used to set another locale."

The Python 3 text model doesn't work properly in the legacy C locale due to the assumption of ASCII as the preferred text encoding, so we run setlocale(LC_ALL, "") early in the startup sequence in order to switch to something more sensible. In Python 3.7+, we're even more opinionated about that, and explicitly coerce the C locale to a UTF-8 based one if there's one available.

If our docs are still saying otherwise anywhere, then our docs are outdated, and need to be fixed.

I understand that the statement "when python starts, it runs using the C locale..." should not be correct anymore (and the doc should then be updated), but in fact this statement is still true on the systems I tested; only, the output of locale.getlocale() at start is in contradiction with the locale really set in fact.

It looks like the setting done by setlocale(LC_ALL, "") at an early stage is lost at some point (only locale.getlocale() seems to "remember" it).

For instance, my box locale is 'fr_FR.UTF-8', so the decimal point is a comma, but when starting python 3.7:


>>> import locale
>>> locale.str(2.4)
'2.4'                     # Wrong: if the locale in use is 'fr_FR.UTF-8', then '2,4' is expected instead
>>> locale.getlocale()
('fr_FR', 'UTF-8')
>>> locale.localeconv()
{'int_curr_symbol': '', 'currency_symbol': '', 'mon_decimal_point': '', 'mon_thousands_sep': '', 'mon_grouping': [], 'positive_sign': '', 'negative_sign': '', 'int_frac_digits': 127, 'frac_digits': 127, 'p_cs_precedes': 127, 'p_sep_by_space': 127, 'n_cs_precedes': 127, 'n_sep_by_space': 127, 'p_sign_posn': 127, 'n_sign_posn': 127, 'decimal_point': '.', 'thousands_sep': '', 'grouping': []}
>>>


Note that the output of localeconv() does match C locale, not 'fr_FR.UTF-8'.

Compare this with the outputs of locale.str() and locale.localeconv() when the locale is explicitly set at start:


>>> import locale
>>> locale.setlocale(locale.LC_ALL, '')
'LC_CTYPE=fr_FR.utf8;LC_NUMERIC=fr_FR.UTF-8;LC_TIME=fr_FR.UTF-8;LC_COLLATE=fr_FR.utf8;LC_MONETARY=fr_FR.UTF-8;LC_MESSAGES=fr_FR.utf8;LC_PAPER=fr_FR.UTF-8;LC_NAME=fr_FR.UTF-8;LC_ADDRESS=fr_FR.UTF-8;LC_TELEPHONE=fr_FR.UTF-8;LC_MEASUREMENT=fr_FR.UTF-8;LC_IDENTIFICATION=fr_FR.UTF-8'
>>> locale.str(2.4)
'2,4'                       # Correct!
>>> locale.localeconv()     # Output of localeconv() does match 'fr_FR.UTF-8' locale
{'int_curr_symbol': 'EUR ', 'currency_symbol': '€', 'mon_decimal_point': ',', 'mon_thousands_sep': '\u202f', 'mon_grouping': [3, 0], 'positive_sign': '', 'negative_sign': '-', 'int_frac_digits': 2, 'frac_digits': 2, 'p_cs_precedes': 0, 'p_sep_by_space': 1, 'n_cs_precedes': 0, 'n_sep_by_space': 1, 'p_sign_posn': 1, 'n_sign_posn': 1, 'decimal_point': ',', 'thousands_sep': '\u202f', 'grouping': [3, 0]}
>>>


Maybe the title of this issue should be turned to "at start, the C locale is in use in spite of locale.getlocale()'s output (python3 on linux)"?




As to the behaviour on Windows, I guess this is another topic (locales belonging to another world on Windows)... but it may be interesting to note that it complies with the current documentation: at start python 3.6 also uses the C locale, and the output of locale.getlocale() is consistent with that. Here is a test on Windows 10:

Python 3.6.3 (v3.6.3:2c5fed8, Oct  3 2017, 18:11:49) [MSC v.1900 64 bit (AMD64)] on win32

>>> import locale
>>> locale.getlocale()
(None, None)
>>> locale.localeconv()
{'decimal_point': '.', 'thousands_sep': '', 'grouping': [], 'int_curr_symbol': '', 'currency_symbol': '', 'mon_decimal_point': '', 'mon_thousands_sep': '', 'mon_grouping': [], 'positive_sign': '', 'negative_sign': '', 'int_frac_digits': 127, 'frac_digits': 127, 'p_cs_precedes': 127, 'p_sep_by_space': 127, 'n_cs_precedes': 127, 'n_sep_by_space': 127, 'p_sign_posn': 127, 'n_sign_posn': 127}
>>> locale.str(2.4)
'2.4'
>>> locale.getdefaultlocale()
('fr_FR', 'cp1252')

When testing this issue, I found a bug in Python :-(

I opened bpo-33954: float.__format__('n') fails with _PyUnicode_CheckConsistency assertion error.

Ah, part of the confusion is that I misremembered the command we run implicitly during startup - it's only `setlocale(LC_CTYPE, "")`, not `setlocale(LC_ALL, "")`.

However, the default category for `locale.getlocale()` is `LC_CTYPE`, so it reports the text encoding locale configured during startup, not the C level default.

The difference on Windows is expected - the startup code that implicitly runs `setlocale(LC_CTYPE, "")` doesn't get compiled in there.

So I think we have a few different potential ways of viewing this bug report:

1. As a docs issue, where we advise users to run `locale.getlocale(locale.LC_MESSAGES)` to find out whether or not a specific locale really has been configured (vs the interpreter's default text encoding change that runs implicitly on startup)
2. As a defaults change for 3.8+, where we switch `locale.getlocale()` over to checking `locale.LC_MESSAGES` instead of `locale.LC_CTYPES`, since the interpreter always sets the latter on startup, so it doesn't convey much useful information.
3. As (1) for maintenance releases, and as (2) for 3.8+