Formatting currencies in the compact format fails for 1000 in locale nl_NL

[bartbroere]

I was formatting monetary amounts in the compact format with babel.numbers.format_compact_currency. Using the Dutch locale (nl_NL) I noticed that it failed for exactly one thousand euros. Add or remove a penny and it works again.

from babel.numbers import format_compact_currency

print(format_compact_currency(1000.01, 'EUR', locale='nl'))
# € 1K

print(format_compact_currency(1000, 'EUR', locale='nl'))
# Traceback (most recent call last):
#   File "/Users/bart/workspace/babel/reproduce_thousand_nl.py", line 6, in <module>
#     print(format_compact_currency(1000, 'EUR', locale='nl'))
#           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
#   File "/Users/bart/workspace/babel/babel/numbers.py", line 924, in format_compact_currency
#     number, format = _get_compact_format(number, compact_format, locale, fraction_digits)
#                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
#   File "/Users/bart/workspace/babel/babel/numbers.py", line 693, in _get_compact_format
#     format = compact_format[plural_form][str(magnitude)]
#              ~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^
#   File "/Users/bart/workspace/babel/babel/localedata.py", line 264, in __getitem__
#     orig = val = self._data[key]
#                  ~~~~~~~~~~^^^^^
# KeyError: '1000'

After reproducing this, I also checked the locale nl, and nl_BE and those have the same issues.

I'm willing to contribute a fix, but people that are more familiar with the codebase might be able to do so quicker.

[bartbroere]

I'm a bit closer to a fix, at least I found a hacky way to avoid the bug, and the right method to debug (babel.numbers._get_compact_format):

diff --git a/babel/numbers.py b/babel/numbers.py
index c4b3c5d..cbcd08b 100644
--- a/babel/numbers.py
+++ b/babel/numbers.py
@@ -690,7 +690,10 @@ def _get_compact_format(
                 plural_form = "other"
             if number == 1 and "1" in compact_format:
                 plural_form = "1"
-            format = compact_format[plural_form][str(magnitude)]
+            if str(magnitude) in compact_format[plural_form]:
+                format = compact_format[plural_form][str(magnitude)]
+            else:
+                format = compact_format["other" if plural_form == "one" else "one"][str(magnitude)]
             number = rounded
             break
     return number, format

The issue is that the plural_form variable is different for Decimal('1000.01') and Decimal('1000') while I suspect it should be the same for those two.

The cldr/cldr-common-47.0/common/main/nl.xml file that is the basis for the nl localisation doesn't have a notation for thousands currency notation, so the K fallback is from somewhere else maybe:

<currencyFormatLength type="short">
	<currencyFormat type="standard">
		<pattern type="1000000" count="one">¤ 0 mln'.'</pattern>
		<pattern type="1000000" count="other">¤ 0 mln'.'</pattern>
		<pattern type="10000000" count="one">¤ 00 mln'.'</pattern>
		<pattern type="10000000" count="other">¤ 00 mln'.'</pattern>
		<pattern type="100000000" count="one">¤ 000 mln'.'</pattern>
		<pattern type="100000000" count="other">¤ 000 mln'.'</pattern>
		<pattern type="1000000000" count="one">¤ 0 mld'.'</pattern>
		<pattern type="1000000000" count="other">¤ 0 mld'.'</pattern>
		<pattern type="10000000000" count="one">¤ 00 mld'.'</pattern>
		<pattern type="10000000000" count="other">¤ 00 mld'.'</pattern>
		<pattern type="100000000000" count="one">¤ 000 mld'.'</pattern>
		<pattern type="100000000000" count="other">¤ 000 mld'.'</pattern>
		<pattern type="1000000000000" count="one">¤ 0 bln'.'</pattern>
		<pattern type="1000000000000" count="other">¤ 0 bln'.'</pattern>
		<pattern type="10000000000000" count="one">¤ 00 bln'.'</pattern>
		<pattern type="10000000000000" count="other">¤ 00 bln'.'</pattern>
		<pattern type="100000000000000" count="one">¤ 000 bln'.'</pattern>
		<pattern type="100000000000000" count="other">¤ 000 bln'.'</pattern>
	</currencyFormat>
</currencyFormatLength>

[bartbroere]

I read here that other is meant to be the implicit default, so I made a first version of a pull request to address this issue. https://github.com/python-babel/babel/blob/master/babel/plural.py#L96-L99.

I also added a smoke test and found out it applies to a few more locales:

FAILED tests/test_smoke.py::test_smoke_numbers[nl_CW] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[nl_BE] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[nl] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[nl_BQ] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[brx] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[nl_SX] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[brx_IN] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[nl_SR] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[nl_AW] - KeyError: '1000'
FAILED tests/test_smoke.py::test_smoke_numbers[nl_NL] - KeyError: '1000'