Compile noninteger TR35 operands to zeroes when emitting Gettext

See http://www.unicode.org/reports/tr35/tr35-numbers.html#Operands
See https://www.gnu.org/software/gettext/manual/html_node/Plural-forms.html

Fixes #287

babel/plural.py

@@ -86,14 +86,14 @@ def __init__(self, rules):
         found = set()
         self.abstract = []
         for key, expr in sorted(list(rules)):
-            if key == 'other':
-                continue
             if key not in _plural_tags:
                 raise ValueError('unknown tag %r' % key)
             elif key in found:
                 raise ValueError('tag %r defined twice' % key)
             found.add(key)
-            self.abstract.append((key, _Parser(expr).ast))
+            ast = _Parser(expr).ast
+            if ast:
+                self.abstract.append((key, ast))
 
     def __repr__(self):
         rules = self.rules
@@ -303,7 +303,7 @@ class RuleError(Exception):
                         .format(_VARS))),
     ('value', re.compile(r'\d+')),
     ('symbol', re.compile(r'%|,|!=|=')),
-    ('ellipsis', re.compile(r'\.\.'))
+    ('ellipsis', re.compile(r'\.{2,3}|\u2026', re.UNICODE))  # U+2026: ELLIPSIS
 ]
 
 
@@ -388,6 +388,11 @@ class _Parser(object):
 
     def __init__(self, string):
         self.tokens = tokenize_rule(string)
+        if not self.tokens:
+            # If the pattern is only samples, it's entirely possible
+            # no stream of tokens whatsoever is generated.
+            self.ast = None
+            return
         self.ast = self.condition()
         if self.tokens:
             raise RuleError('Expected end of rule, got %r' %
@@ -480,6 +485,9 @@ def _unary_compiler(tmpl):
     return lambda self, x: tmpl % self.compile(x)
 
 
+compile_zero = lambda x: '0'
+
+
 class _Compiler(object):
     """The compilers are able to transform the expressions into multiple
     output formats.
@@ -526,6 +534,12 @@ def compile_relation(self, method, expr, range_list):
 class _GettextCompiler(_Compiler):
     """Compile into a gettext plural expression."""
 
+    compile_i = _Compiler.compile_n
+    compile_v = compile_zero
+    compile_w = compile_zero
+    compile_f = compile_zero
+    compile_t = compile_zero
+
     def compile_relation(self, method, expr, range_list):
         rv = []
         expr = self.compile(expr)
@@ -552,10 +566,10 @@ class _JavaScriptCompiler(_GettextCompiler):
     # XXX: presently javascript does not support any of the
     # fraction support and basically only deals with integers.
     compile_i = lambda x: 'parseInt(n, 10)'
-    compile_v = lambda x: '0'
-    compile_w = lambda x: '0'
-    compile_f = lambda x: '0'
-    compile_t = lambda x: '0'
+    compile_v = compile_zero
+    compile_w = compile_zero
+    compile_f = compile_zero
+    compile_t = compile_zero
 
     def compile_relation(self, method, expr, range_list):
         code = _GettextCompiler.compile_relation(

tests/test_plural.py

@@ -13,7 +13,7 @@
 import unittest
 import pytest
 
-from babel import plural
+from babel import plural, localedata
 from babel._compat import Decimal
 
 
@@ -254,3 +254,17 @@ def test_extract_operands(source, n, i, v, w, f, t):
     source = Decimal(source) if isinstance(source, str) else source
     assert (plural.extract_operands(source) ==
             Decimal(n), i, v, w, f, t)
+
+
+@pytest.mark.parametrize('locale', ('ru', 'pl'))
+def test_gettext_compilation(locale):
+    # Test that new plural form elements introduced in recent CLDR versions
+    # are compiled "down" to `n` when emitting Gettext rules.
+    ru_rules = localedata.load(locale)['plural_form'].rules
+    chars = 'ivwft'
+    # Test that these rules are valid for this test; i.e. that they contain at least one
+    # of the gettext-unsupported characters.
+    assert any((" " + ch + " ") in rule for ch in chars for rule in ru_rules.values())
+    # Then test that the generated value indeed does not contain these.
+    ru_rules_gettext = plural.to_gettext(ru_rules)
+    assert not any(ch in ru_rules_gettext for ch in chars)