locale: wrong thousand separator for "fr_CH.UTF-8"

[sphakka]

Bug report

Bug description:

With locale "fr_CH.UTF-8", the thousand separator is (possibly) wrong. It should be a plain apostrophe (U+0027) or a hard space (U+00A0 I guess), not a modifier letter apostrophe (U+02BC, or whatever it is -- looks weird) -- cf, page 35 of the official style guide.

Example (with workaround) -- numbers from a financial CSV sheet issued by UBS Switzerland:

py>> import locale
py>> locale.setlocale(locale.LC_NUMERIC, 'fr_CH.utf8')
'fr_CH.utf8'
py>> locale.localeconv()["thousands_sep"]
'’'
py>> locale.atof("1'080.10")
Traceback (most recent call last):
  File "<python-input-3>", line 1, in <module>
    locale.atof("1'080.10")
    ~~~~~~~~~~~^^^^^^^^^^^^
  File "/usr/lib/python3.14/locale.py", line 328, in atof
    return func(delocalize(string))
ValueError: could not convert string to float: "1'080.10"
py>> locale._override_localeconv={"thousands_sep":"'"}
py>> locale.atof("1'080.10")
1080.1

Ideally, both separator options should be supported, though that might make for another bug report. Let me know.

Tested with Python-3.13 and Python-3.14.

CPython versions tested on:

3.14

Operating systems tested on:

Linux

[StanFromIreland]

What does LC_NUMERIC=fr_CH.utf8 locale -ck LC_NUMERIC return for you?

On my system, the locale data is correct:

$ LC_NUMERIC=fr_CH.utf8 locale -ck LC_NUMERIC
LC_NUMERIC
decimal_point="."
thousands_sep="’"
grouping=3
numeric-decimal-point-wc=46
numeric-thousands-sep-wc=8217
numeric-codeset="UTF-8"

[sphakka]

LC_NUMERIC=fr_CH.utf8 locale -ck LC_NUMERIC

It's exactly the same 🤔. Upstream issue (locale-gen)?

[picnixz]

Yes I think so. IIRC, we're just using the separator given by the locale.

[StanFromIreland]

We don't provide locale data, we use the system data.

[picnixz]

Note that the guidelines given by the country might not be aligned with the locale specifications themselves. I don't know where fr_CH.UTF-8 is actually specified but it might be coming from other specs (that are independent of the link you provide)

[StanFromIreland]

This is not an issue for this tracker, if you wish to change this I suggest you report to the glibc (or whoever provides your locale data) locale data maintainers.

[sphakka]

Thanks for the info. I'll dig upstream.

[picnixz]

I'm actually surprised by the error actually:

def delocalize(string):
    "Parses a string as a normalized number according to the locale settings."

    conv = localeconv()

    #First, get rid of the grouping
    ts = conv['thousands_sep']
    if ts:
        string = string.replace(ts, '')

    #next, replace the decimal point with a dot
    dd = conv['decimal_point']
    if dd:
        string = string.replace(dd, '.')
    return string

So delocalize should actually get rid of this weird sep.