Fixed #17375 -- Changed makemessages to use xgettext with --files-from

Changed the way makemessages invokes xgettext from one call per
translatable file to one call per locale directory (using --files-from).
This allows to avoid https://savannah.gnu.org/bugs/index.php?35027 and,
as a positive side effect, speeds up localization build.

django/core/management/commands/makemessages.py

@@ -11,6 +11,7 @@ from itertools import dropwhile
 
 import django
 from django.conf import settings
+from django.core.files.temp import NamedTemporaryFile
 from django.core.management.base import BaseCommand, CommandError
 from django.core.management.utils import (
     find_command, handle_extensions, popen_wrapper,
@@ -74,103 +75,96 @@ class TranslatableFile(object):
     def path(self):
         return os.path.join(self.dirpath, self.file)
 
-    def process(self, command, domain):
-        """
-        Extract translatable literals from self.file for :param domain:,
-        creating or updating the POT file.
 
-        Uses the xgettext GNU gettext utility.
+class BuildFile(object):
+    """
+    Represents the state of a translatable file during the build process.
+    """
+    def __init__(self, command, domain, translatable):
+        self.command = command
+        self.domain = domain
+        self.translatable = translatable
+
+    @cached_property
+    def is_templatized(self):
+        if self.domain == 'djangojs':
+            return self.command.gettext_version < (0, 18, 3)
+        elif self.domain == 'django':
+            file_ext = os.path.splitext(self.translatable.file)[1]
+            return file_ext != '.py'
+        return False
+
+    @cached_property
+    def path(self):
+        return self.translatable.path
+
+    @cached_property
+    def work_path(self):
+        """
+        Path to a file which is being fed into GNU gettext pipeline. This may
+        be either a translatable or its preprocessed version.
+        """
+        if not self.is_templatized:
+            return self.path
+        extension = {
+            'djangojs': 'c',
+            'django': 'py',
+        }.get(self.domain)
+        filename = '%s.%s' % (self.translatable.file, extension)
+        return os.path.join(self.translatable.dirpath, filename)
+
+    def preprocess(self):
+        """
+        Preprocess (if necessary) a translatable file before passing it to
+        xgettext GNU gettext utility.
         """
         from django.utils.translation import templatize
 
-        if command.verbosity > 1:
+        if not self.is_templatized:
-            command.stdout.write('processing file %s in %s\n' % (self.file, self.dirpath))
-        file_ext = os.path.splitext(self.file)[1]
-        if domain == 'djangojs':
-            orig_file = os.path.join(self.dirpath, self.file)
-            work_file = orig_file
-            is_templatized = command.gettext_version < (0, 18, 3)
-            if is_templatized:
-                with io.open(orig_file, 'r', encoding=settings.FILE_CHARSET) as fp:
-                    src_data = fp.read()
-                src_data = prepare_js_for_gettext(src_data)
-                work_file = os.path.join(self.dirpath, '%s.c' % self.file)
-                with io.open(work_file, "w", encoding='utf-8') as fp:
-                    fp.write(src_data)
-            args = [
-                'xgettext',
-                '-d', domain,
-                '--language=%s' % ('C' if is_templatized else 'JavaScript',),
-                '--keyword=gettext_noop',
-                '--keyword=gettext_lazy',
-                '--keyword=ngettext_lazy:1,2',
-                '--keyword=pgettext:1c,2',
-                '--keyword=npgettext:1c,2,3',
-                '--output=-'
-            ] + command.xgettext_options
-            args.append(work_file)
-        elif domain == 'django':
-            orig_file = os.path.join(self.dirpath, self.file)
-            work_file = orig_file
-            is_templatized = file_ext != '.py'
-            if is_templatized:
-                with io.open(orig_file, encoding=settings.FILE_CHARSET) as fp:
-                    src_data = fp.read()
-                content = templatize(src_data, orig_file[2:])
-                work_file = os.path.join(self.dirpath, '%s.py' % self.file)
-                with io.open(work_file, "w", encoding='utf-8') as fp:
-                    fp.write(content)
-            args = [
-                'xgettext',
-                '-d', domain,
-                '--language=Python',
-                '--keyword=gettext_noop',
-                '--keyword=gettext_lazy',
-                '--keyword=ngettext_lazy:1,2',
-                '--keyword=ugettext_noop',
-                '--keyword=ugettext_lazy',
-                '--keyword=ungettext_lazy:1,2',
-                '--keyword=pgettext:1c,2',
-                '--keyword=npgettext:1c,2,3',
-                '--keyword=pgettext_lazy:1c,2',
-                '--keyword=npgettext_lazy:1c,2,3',
-                '--output=-'
-            ] + command.xgettext_options
-            args.append(work_file)
-        else:
             return
-        msgs, errors, status = gettext_popen_wrapper(args)
-        if errors:
-            if status != STATUS_OK:
-                if is_templatized:
-                    os.unlink(work_file)
-                raise CommandError(
-                    "errors happened while running xgettext on %s\n%s" %
-                    (self.file, errors))
-            elif command.verbosity > 0:
-                # Print warnings
-                command.stdout.write(errors)
-        if msgs:
-            # Write/append messages to pot file
-            if self.locale_dir is NO_LOCALE_DIR:
-                file_path = os.path.normpath(os.path.join(self.dirpath, self.file))
-                raise CommandError(
-                    "Unable to find a locale path to store translations for file %s" % file_path)
-            potfile = os.path.join(self.locale_dir, '%s.pot' % str(domain))
-            if is_templatized:
-                # Remove '.py' suffix
-                if os.name == 'nt':
-                    # Preserve '.\' prefix on Windows to respect gettext behavior
-                    old = '#: ' + work_file
-                    new = '#: ' + orig_file
-                else:
-                    old = '#: ' + work_file[2:]
-                    new = '#: ' + orig_file[2:]
-                msgs = msgs.replace(old, new)
-            write_pot_file(potfile, msgs)
 
-        if is_templatized:
+        with io.open(self.path, 'r', encoding=settings.FILE_CHARSET) as fp:
-            os.unlink(work_file)
+            src_data = fp.read()
+
+        if self.domain == 'djangojs':
+            content = prepare_js_for_gettext(src_data)
+        elif self.domain == 'django':
+            content = templatize(src_data, self.path[2:])
+
+        with io.open(self.work_path, 'w', encoding='utf-8') as fp:
+            fp.write(content)
+
+    def postprocess_messages(self, msgs):
+        """
+        Postprocess messages generated by xgettext GNU gettext utility.
+
+        Transform paths as if these messages were generated from original
+        translatable files rather than from preprocessed versions.
+        """
+        if not self.is_templatized:
+            return msgs
+
+        # Remove '.py' suffix
+        if os.name == 'nt':
+            # Preserve '.\' prefix on Windows to respect gettext behavior
+            old = '#: ' + self.work_path
+            new = '#: ' + self.path
+        else:
+            old = '#: ' + self.work_path[2:]
+            new = '#: ' + self.path[2:]
+
+        return msgs.replace(old, new)
+
+    def cleanup(self):
+        """
+        Remove a preprocessed copy of a translatable file (if any).
+        """
+        if self.is_templatized:
+            # This check is needed for the case of a symlinked file and its
+            # source being processed inside a single group (locale dir);
+            # removing either of those two removes both.
+            if os.path.exists(self.work_path):
+                os.unlink(self.work_path)
 
 
 def write_pot_file(potfile, msgs):
@@ -194,6 +188,9 @@ class Command(BaseCommand):
 "applications) directory.\n\nYou must run this command with one of either the "
 "--locale, --exclude or --all options.")
 
+    translatable_file_class = TranslatableFile
+    build_file_class = BuildFile
+
     requires_system_checks = False
     leave_locale_alone = True
 
@@ -353,18 +350,7 @@ class Command(BaseCommand):
         """
         file_list = self.find_files(".")
         self.remove_potfiles()
-        for f in file_list:
+        self.process_files(file_list)
-            try:
-                f.process(self, self.domain)
-            except UnicodeDecodeError as e:
-                self.stdout.write(
-                    "UnicodeDecodeError: skipped file %s in %s (reason: %s)" % (
-                        f.file,
-                        f.dirpath,
-                        e,
-                    )
-                )
-
         potfiles = []
         for path in self.locale_paths:
             potfile = os.path.join(path, '%s.pot' % str(self.domain))
@@ -443,9 +429,116 @@ class Command(BaseCommand):
                         locale_dir = self.default_locale_path
                     if not locale_dir:
                         locale_dir = NO_LOCALE_DIR
-                    all_files.append(TranslatableFile(dirpath, filename, locale_dir))
+                    all_files.append(self.translatable_file_class(dirpath, filename, locale_dir))
         return sorted(all_files)
 
+    def process_files(self, file_list):
+        """
+        Group translatable files by locale directory and run pot file build
+        process for each group.
+        """
+        file_groups = {}
+        for translatable in file_list:
+            file_group = file_groups.setdefault(translatable.locale_dir, [])
+            file_group.append(translatable)
+        for locale_dir, files in file_groups.items():
+            self.process_locale_dir(locale_dir, files)
+
+    def process_locale_dir(self, locale_dir, files):
+        """
+        Extract translatable literals from the specified files, creating or
+        updating the POT file for a given locale directory.
+
+        Uses the xgettext GNU gettext utility.
+        """
+        build_files = []
+        for translatable in files:
+            if self.verbosity > 1:
+                self.stdout.write('processing file %s in %s\n' % (
+                    translatable.file, translatable.dirpath
+                ))
+            if self.domain not in ('djangojs', 'django'):
+                continue
+            build_file = self.build_file_class(self, self.domain, translatable)
+            try:
+                build_file.preprocess()
+            except UnicodeDecodeError as e:
+                self.stdout.write(
+                    'UnicodeDecodeError: skipped file %s in %s (reason: %s)' % (
+                        translatable.file, translatable.dirpath, e,
+                    )
+                )
+                continue
+            build_files.append(build_file)
+
+        if self.domain == 'djangojs':
+            is_templatized = build_file.is_templatized
+            args = [
+                'xgettext',
+                '-d', self.domain,
+                '--language=%s' % ('C' if is_templatized else 'JavaScript',),
+                '--keyword=gettext_noop',
+                '--keyword=gettext_lazy',
+                '--keyword=ngettext_lazy:1,2',
+                '--keyword=pgettext:1c,2',
+                '--keyword=npgettext:1c,2,3',
+                '--output=-',
+            ]
+        elif self.domain == 'django':
+            args = [
+                'xgettext',
+                '-d', self.domain,
+                '--language=Python',
+                '--keyword=gettext_noop',
+                '--keyword=gettext_lazy',
+                '--keyword=ngettext_lazy:1,2',
+                '--keyword=ugettext_noop',
+                '--keyword=ugettext_lazy',
+                '--keyword=ungettext_lazy:1,2',
+                '--keyword=pgettext:1c,2',
+                '--keyword=npgettext:1c,2,3',
+                '--keyword=pgettext_lazy:1c,2',
+                '--keyword=npgettext_lazy:1c,2,3',
+                '--output=-',
+            ]
+        else:
+            return
+
+        input_files = [bf.work_path for bf in build_files]
+        with NamedTemporaryFile(mode='w+') as input_files_list:
+            input_files_list.write('\n'.join(input_files))
+            input_files_list.flush()
+            args.extend(['--files-from', input_files_list.name])
+            args.extend(self.xgettext_options)
+            msgs, errors, status = gettext_popen_wrapper(args)
+
+        if errors:
+            if status != STATUS_OK:
+                for build_file in build_files:
+                    build_file.cleanup()
+                raise CommandError(
+                    'errors happened while running xgettext on %s\n%s' %
+                    ('\n'.join(input_files), errors)
+                )
+            elif self.verbosity > 0:
+                # Print warnings
+                self.stdout.write(errors)
+
+        if msgs:
+            if locale_dir is NO_LOCALE_DIR:
+                file_path = os.path.normpath(build_files[0].path)
+                raise CommandError(
+                    'Unable to find a locale path to store translations for '
+                    'file %s' % file_path
+                )
+            for build_file in build_files:
+                msgs = build_file.postprocess_messages(msgs)
+            potfile = os.path.join(locale_dir, '%s.pot' % str(self.domain))
+            write_pot_file(potfile, msgs)
+
+        for build_file in build_files:
+            build_file.cleanup()
+
     def write_po_file(self, potfile, locale):
         """
         Creates or updates the PO file for self.domain and :param locale:.

docs/releases/1.9.txt

@@ -380,6 +380,9 @@ Internationalization
   project and it will find all the app message files that were created by
   :djadmin:`makemessages`.
 
+* :djadmin:`makemessages` now calls xgettext once per locale directory rather
+  than once per translatable file. This speeds up localization builds.
+
 * :ttag:`blocktrans` supports assigning its output to a variable using
   ``asvar``.
 

tests/i18n/commands/templates/plural.djtpl

@@ -4,3 +4,5 @@ This file has a literal with plural forms. When processed first, makemessages
 shouldn't create a .po file with duplicate `Plural-Forms` headers
 {% endcomment %}
 {% blocktrans count number=3 %}{{ number }} Bar{% plural %}{{ number }} Bars{% endblocktrans %}
+
+{% trans 'First `trans`, then `blocktrans` with a plural' %}

tests/i18n/commands/templates/test.html

@@ -80,3 +80,9 @@ continued here.{% endcomment %}
   should be trimmed.
 {% endblocktrans %}
 {% trans "I'm on line 82" %}
+
+{% blocktrans trimmed count counter=mylist|length %}
+First `trans`, then `blocktrans` with a plural
+{% plural %}
+Plural for a `trans` and `blocktrans` collision case
+{% endblocktrans %}

tests/i18n/test_extraction.py

@@ -80,6 +80,9 @@ class ExtractorTests(SimpleTestCase):
     def assertMsgId(self, msgid, haystack, use_quotes=True):
         return self._assertPoKeyword('msgid', msgid, haystack, use_quotes=use_quotes)
 
+    def assertMsgIdPlural(self, msgid, haystack, use_quotes=True):
+        return self._assertPoKeyword('msgid_plural', msgid, haystack, use_quotes=use_quotes)
+
     def assertMsgStr(self, msgstr, haystack, use_quotes=True):
         return self._assertPoKeyword('msgstr', msgstr, haystack, use_quotes=use_quotes)
 
@@ -525,6 +528,21 @@ class CopyPluralFormsExtractorTests(ExtractorTests):
             found = re.findall(r'^(?P<value>"Plural-Forms.+?\\n")\s*$', po_contents, re.MULTILINE | re.DOTALL)
             self.assertEqual(1, len(found))
 
+    def test_trans_and_plural_blocktrans_collision(self):
+        """
+        Ensures a correct workaround for the gettext bug when handling a literal
+        found inside a {% trans %} tag and also in another file inside a
+        {% blocktrans %} with a plural (#17375).
+        """
+        os.chdir(self.test_dir)
+        management.call_command('makemessages', locale=[LOCALE], extensions=['html', 'djtpl'], verbosity=0)
+        self.assertTrue(os.path.exists(self.PO_FILE))
+        with open(self.PO_FILE, 'r') as fp:
+            po_contents = force_text(fp.read())
+            self.assertNotIn("#-#-#-#-#  django.pot (PACKAGE VERSION)  #-#-#-#-#\\n", po_contents)
+            self.assertMsgId('First `trans`, then `blocktrans` with a plural', po_contents)
+            self.assertMsgIdPlural('Plural for a `trans` and `blocktrans` collision case', po_contents)
+
 
 class NoWrapExtractorTests(ExtractorTests):