Fixed #6409 -- Unbreak compound locale name parsing (e.g. zh-cn).

This was inadvertently broken back in [6608]. Slightly backwards-incompatible:
people specifying "es_AR" in their LANGUAGES list will need to change that to
"es-ar". Thanks, simonb and Ramiro Morales for making the effort to fix this.


git-svn-id: http://code.djangoproject.com/svn/django/trunk@7091 bcc190cf-cafb-0310-a4f2-bffc1f526a37

django/conf/global_settings.py

@@ -47,7 +47,7 @@ LANGUAGES = (
     ('el', gettext_noop('Greek')),
     ('en', gettext_noop('English')),
     ('es', gettext_noop('Spanish')),
-    ('es_AR', gettext_noop('Argentinean Spanish')),
+    ('es-ar', gettext_noop('Argentinean Spanish')),
     ('fa', gettext_noop('Persian')),
     ('fi', gettext_noop('Finnish')),
     ('fr', gettext_noop('French')),

django/utils/translation/trans_real.py

@@ -42,7 +42,10 @@ accept_language_re = re.compile(r'''
         ''', re.VERBOSE)
 
 def to_locale(language, to_lower=False):
-    "Turns a language name (en-us) into a locale name (en_US)."
+    """
+    Turns a language name (en-us) into a locale name (en_US). If 'to_lower' is
+    True, the last component is lower-cased (en_us).
+    """
     p = language.find('-')
     if p >= 0:
         if to_lower:
@@ -357,19 +360,20 @@ def get_language_from_request(request):
         return lang_code
 
     accept = request.META.get('HTTP_ACCEPT_LANGUAGE', '')
-    for lang, unused in parse_accept_lang_header(accept):
+    for accept_lang, unused in parse_accept_lang_header(accept):
-        if lang == '*':
+        if accept_lang == '*':
             break
 
         # We have a very restricted form for our language files (no encoding
         # specifier, since they all must be UTF-8 and only one possible
         # language each time. So we avoid the overhead of gettext.find() and
-        # look up the MO file manually.
+        # work out the MO file manually.
 
-        normalized = locale.locale_alias.get(to_locale(lang, True))
+        # 'normalized' is the root name of the locale in POSIX format (which is
+        # the format used for the directories holding the MO files).
+        normalized = locale.locale_alias.get(to_locale(accept_lang, True))
         if not normalized:
             continue
-
         # Remove the default encoding from locale_alias
         normalized = normalized.split('.')[0]
 
@@ -378,10 +382,11 @@ def get_language_from_request(request):
             # need to check again.
             return _accepted[normalized]
 
-        for lang in (normalized, normalized.split('_')[0]):
+        for lang, dirname in ((accept_lang, normalized),
+                (accept_lang.split('-')[0], normalized.split('_')[0])):
             if lang not in supported:
                 continue
-            langfile = os.path.join(globalpath, lang, 'LC_MESSAGES',
+            langfile = os.path.join(globalpath, dirname, 'LC_MESSAGES',
                     'django.mo')
             if os.path.exists(langfile):
                 _accepted[normalized] = lang

tests/regressiontests/i18n/misc.py

@@ -2,6 +2,11 @@ tests = """
 >>> from django.utils.translation.trans_real import parse_accept_lang_header
 >>> p = parse_accept_lang_header
 
+#
+# Testing HTTP header parsing. First, we test that we can parse the values
+# according to the spec (and that we extract all the pieces in the right order).
+#
+
 Good headers.
 >>> p('de')
 [('de', 1.0)]
@@ -54,4 +59,44 @@ Bad headers; should always return [].
 >>> p('')
 []
 
+#
+# Now test that we parse a literal HTTP header correctly.
+#
+
+>>> from django.utils.translation.trans_real import get_language_from_request
+>>> g = get_language_from_request
+>>> from django.http import HttpRequest
+>>> r = HttpRequest
+>>> r.COOKIES = {}
+
+These tests assumes the es, es_AR, pt and pt_BR translations exit in the Django
+source tree.
+>>> r.META = {'HTTP_ACCEPT_LANGUAGE': 'pt-br'}
+>>> g(r)
+'pt-br'
+>>> r.META = {'HTTP_ACCEPT_LANGUAGE': 'pt'}
+>>> g(r)
+'pt'
+>>> r.META = {'HTTP_ACCEPT_LANGUAGE': 'es,de'}
+>>> g(r)
+'es'
+>>> r.META = {'HTTP_ACCEPT_LANGUAGE': 'es-ar,de'}
+>>> g(r)
+'es-ar'
+
+This test assumes there won't be a Django translation to a US variation
+of the Spanish language, a safe assumption. When the user sets it
+as the preferred language, the main 'es' translation should be selected
+instead.
+>>> r.META = {'HTTP_ACCEPT_LANGUAGE': 'es-us'}
+>>> g(r)
+'es'
+
+This tests the following scenario: there isn't a main language (zh)
+translation of Django but there is a translation to variation (zh_CN)
+the user sets zh-cn as the preferred language, it should be selected by
+Django without falling back nor ignoring it.
+>>> r.META = {'HTTP_ACCEPT_LANGUAGE': 'zh-cn,de'}
+>>> g(r)
+'zh-cn'
 """