Add support for compact decimal formats

[DenverCoder1]

Adds support for compact decimal formats.

Implementation based on details found at https://www.unicode.org/reports/tr35/tr35-45/tr35-numbers.html#Compact_Number_Formats

    >>> format_compact_decimal(12345, format_type="short", locale='en_US')
    u'12K'
    >>> format_compact_decimal(12345, format_type="long", locale='en_US')
    u'12 thousand'
    >>> format_compact_decimal(12345, format_type="short", locale='en_US', fraction_digits=2)
    u'12.35K'
    >>> format_compact_decimal(1234567, format_type="short", locale="ja_JP")
    u'123万'
    >>> format_compact_decimal(2345678, format_type="long", locale="mk")
    u'2 милиони'
    >>> format_compact_decimal(21098765, format_type="long", locale="mk")
    u'21 милион'

Closes #908

[jun66j5]

If the given number is negative, the formatted text is not compact.

>>> numbers.format_decimal(123456789, compact='short', locale='en_US')
'123M'
>>> numbers.format_decimal(123456789, compact='short', locale='ja')
'1億'
>>> numbers.format_decimal(-123456789, compact='short', locale='en_US')
'-123,456,789'
>>> numbers.format_decimal(-123456789, compact='short', locale='ja')
'-123,456,789'

[jun66j5]

Several locales have different formats between count="one" and count="other" (e.g. mk, ps, ...). I think the count="one" pattern should be handled.

>>> from babel.core import Locale
>>> from babel.localedata import locale_identifiers
>>> def show_different_formats():
...   for identifier in sorted(locale_identifiers()):
...     locale = Locale.parse(identifier)
...     fmts = locale.compact_decimal_formats['short']
...     if 'one' in fmts and 'other' in fmts:
...       for s in sorted(set(fmts['one']) & set(fmts['other'])):
...         fmt_one = fmts['one'][s]
...         fmt_other = fmts['other'][s]
...         if fmt_one.pattern != fmt_other.pattern:
...           print(identifier, repr(s), repr(fmt_one.pattern), repr(fmt_other.pattern))
...
>>> show_different_formats()
bn '10000000000' '00\xa0শত\xa0কো' '00শত\xa0কো'
bn_BD '10000000000' '00\xa0শত\xa0কো' '00শত\xa0কো'
bn_IN '10000000000' '00\xa0শত\xa0কো' '00শত\xa0কো'
mk '100000000000' "000\xa0мј'.'" "000\xa0ми'.'"
mk_MK '100000000000' "000\xa0мј'.'" "000\xa0ми'.'"
ps '100000000000' '000G' '000B'
ps_AF '100000000000' '000G' '000B'
ps_PK '100000000000' '000G' '000B'
sw '1000000' '0M;-0M' '0M'
sw '10000000' '00M;-00M' '00M'
sw '100000000' '000M;-000M' '000M'
sw '1000000000000' '0T;-0T' '0T'
sw '10000000000000' '00T;-00T' '00T'
sw '100000000000000' '000T;-000T' '000T'
sw_CD '1000000' '0M;-0M' '0M'
sw_CD '10000000' '00M;-00M' '00M'
sw_CD '100000000' '000M;-000M' '000M'
sw_CD '1000000000000' '0T;-0T' '0T'
sw_CD '10000000000000' '00T;-00T' '00T'
sw_CD '100000000000000' '000T;-000T' '000T'
sw_TZ '1000000' '0M;-0M' '0M'
sw_TZ '10000000' '00M;-00M' '00M'
sw_TZ '100000000' '000M;-000M' '000M'
sw_TZ '1000000000000' '0T;-0T' '0T'
sw_TZ '10000000000000' '00T;-00T' '00T'
sw_TZ '100000000000000' '000T;-000T' '000T'
sw_UG '1000000' '0M;-0M' '0M'
sw_UG '10000000' '00M;-00M' '00M'
sw_UG '100000000' '000M;-000M' '000M'
sw_UG '1000000000000' '0T;-0T' '0T'
sw_UG '10000000000000' '00T;-00T' '00T'
sw_UG '100000000000000' '000T;-000T' '000T'

[DenverCoder1]

Several locales have different formats between count="one" and count="other" (e.g. mk, ps, ...). I think the count="one" pattern should be handled.

Unless I understood incorrectly, I believe it does already when number is 1 after dividing.

[DenverCoder1]

In the latest commit, the parameter compact was renamed to format_type to make it more similar to the parameters to format_currency.

format_type can be None to use the format locale.decimal_formats[None], or "short" / "long" to use the respective compact decimal format.

(Note: compact format currencies are not implemented by this PR, they appear to partially implemented already and broken and can be corrected in a later PR).

[jun66j5]

Unless I understood incorrectly, I believe it does already when number is 1 after dividing.

No. We should determine to use which "count" attribute using plural form of the locale.

>>> from babel.core import Locale
>>> mk = Locale.parse('mk')
>>> mk.plural_form
<PluralRule 'one: v in 0 and i mod 10 in 1 and i mod 100 not in 11 or f mod 10 in 1 and f mod 100 not in 11'>
>>> for n in (0, 1, 2, 11, 21, 100, 101):
...   n, mk.plural_form(n)
...
(0, 'other')
(1, 'one')
(2, 'other')
(11, 'other')
(21, 'one')
(100, 'other')
(101, 'one')

[DenverCoder1]

Thanks, that actually makes a lot more sense with the examples. I believe it still should check after dividing, though, since the plural rule seems to only apply to the first few digits.

[jun66j5]

I think we could add doc tests for the plural form like:

index 7e0ddd63a..fc2164183 100644
--- a/babel/numbers.py
+++ b/babel/numbers.py
@@ -417,6 +417,10 @@ def format_decimal(
     u'12.35K'
     >>> format_decimal(1234567, locale='ja_JP', format_type="short")
     u'123万'
+    >>> format_decimal(2345678, locale='mk', format_type='long')
+    u'2 милиони'
+    >>> format_decimal(21098765, locale='mk', format_type='long')
+    u'21 милион'

     :param number: the number to format
     :param format:

[DenverCoder1]

I think we could add doc tests for the plural form

Added 👍

[codecov]

Codecov Report

Merging #909 (6e768be) into master (cdc77b4) will decrease coverage by 0.31%.
The diff coverage is 100.00%.

❗ Current head 6e768be differs from pull request most recent head 6a29607. Consider uploading reports for the commit 6a29607 to get more accurate results

@@            Coverage Diff             @@
##           master     #909      +/-   ##
==========================================
- Coverage   91.33%   91.01%   -0.32%     
==========================================
  Files          21       21              
  Lines        4176     4198      +22     
==========================================
+ Hits         3814     3821       +7     
- Misses        362      377      +15     

Impacted Files	Coverage Δ
babel/core.py	96.37% <100.00%> (+0.03%)	⬆️
babel/numbers.py	98.12% <100.00%> (+0.12%)	⬆️
babel/localtime/_win32.py	21.56% <0.00%> (-39.22%)	⬇️
babel/localtime/__init__.py	61.53% <0.00%> (-2.57%)	⬇️
babel/messages/frontend.py	87.17% <0.00%> (-0.75%)	⬇️
babel/localedata.py	95.86% <0.00%> (ø)
babel/messages/jslexer.py	87.00% <0.00%> (+0.33%)	⬆️
babel/localtime/_unix.py	24.13% <0.00%> (+0.52%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[akx]

Thank you for getting to work on this! The code looks mainly great, just a couple thoughts here and there, and maybe we should discuss whether to overload the format_decimal API further.

(Also thanks, @jun66j5, for thoroughly reviewing this earlier!)

[akx]

Looks quite excellent! Just a couple of tiny comments to go.

[akx]

Great, thank you!