285 lines
9.7 KiB
Python
Executable file
285 lines
9.7 KiB
Python
Executable file
#!/bin/python
|
|
|
|
import enum
|
|
import fnmatch
|
|
import os
|
|
import re
|
|
import shutil
|
|
import subprocess
|
|
import sys
|
|
|
|
|
|
class Message:
|
|
__slots__ = ("msgid", "msgid_plural", "msgctxt", "comments", "locations")
|
|
|
|
def format(self):
|
|
lines = []
|
|
|
|
if self.comments:
|
|
for i, content in enumerate(self.comments):
|
|
prefix = "#. TRANSLATORS:" if i == 0 else "#."
|
|
lines.append(prefix + content)
|
|
|
|
lines.append("#: " + " ".join(self.locations))
|
|
|
|
if self.msgctxt:
|
|
lines.append('msgctxt "{}"'.format(self.msgctxt))
|
|
|
|
if self.msgid_plural:
|
|
lines += [
|
|
'msgid "{}"'.format(self.msgid),
|
|
'msgid_plural "{}"'.format(self.msgid_plural),
|
|
'msgstr[0] ""',
|
|
'msgstr[1] ""',
|
|
]
|
|
else:
|
|
lines += [
|
|
'msgid "{}"'.format(self.msgid),
|
|
'msgstr ""',
|
|
]
|
|
|
|
return "\n".join(lines)
|
|
|
|
|
|
messages_map = {} # (id, context) -> Message.
|
|
|
|
line_nb = False
|
|
|
|
for arg in sys.argv[1:]:
|
|
if arg == "--with-line-nb":
|
|
print("Enabling line numbers in the context locations.")
|
|
line_nb = True
|
|
else:
|
|
os.sys.exit("Non supported argument '" + arg + "'. Aborting.")
|
|
|
|
|
|
if not os.path.exists("editor"):
|
|
os.sys.exit("ERROR: This script should be started from the root of the git repo.")
|
|
|
|
|
|
matches = []
|
|
for root, dirnames, filenames in os.walk("."):
|
|
dirnames[:] = [d for d in dirnames if d not in ["thirdparty"]]
|
|
for filename in fnmatch.filter(filenames, "*.cpp"):
|
|
matches.append(os.path.join(root, filename))
|
|
for filename in fnmatch.filter(filenames, "*.h"):
|
|
matches.append(os.path.join(root, filename))
|
|
matches.sort()
|
|
|
|
|
|
remaps = {}
|
|
remap_re = re.compile(r'^\t*capitalize_string_remaps\["(?P<from>.+)"\] = (String::utf8\()?"(?P<to>.+)"')
|
|
with open("editor/editor_property_name_processor.cpp") as f:
|
|
for line in f:
|
|
m = remap_re.search(line)
|
|
if m:
|
|
remaps[m.group("from")] = m.group("to")
|
|
|
|
|
|
main_po = """
|
|
# LANGUAGE translation of the Godot Engine editor.
|
|
# Copyright (c) 2007-2022 Juan Linietsky, Ariel Manzur.
|
|
# Copyright (c) 2014-2022 Godot Engine contributors (cf. AUTHORS.md).
|
|
# This file is distributed under the same license as the Godot source code.
|
|
#
|
|
# FIRST AUTHOR <EMAIL@ADDRESS>, YEAR.
|
|
#
|
|
#, fuzzy
|
|
msgid ""
|
|
msgstr ""
|
|
"Project-Id-Version: Godot Engine editor\\n"
|
|
"Report-Msgid-Bugs-To: https://github.com/godotengine/godot\\n"
|
|
"MIME-Version: 1.0\\n"
|
|
"Content-Type: text/plain; charset=UTF-8\\n"
|
|
"Content-Transfer-Encoding: 8-bit\\n"\n
|
|
"""
|
|
|
|
|
|
class ExtractType(enum.IntEnum):
|
|
TEXT = 1
|
|
PROPERTY_PATH = 2
|
|
GROUP = 3
|
|
|
|
|
|
# Regex "(?P<name>([^"\\]|\\.)*)" creates a group named `name` that matches a string.
|
|
message_patterns = {
|
|
re.compile(r'RTR\("(?P<message>([^"\\]|\\.)*)"(, "(?P<context>([^"\\]|\\.)*)")?\)'): ExtractType.TEXT,
|
|
re.compile(r'TTR\("(?P<message>([^"\\]|\\.)*)"(, "(?P<context>([^"\\]|\\.)*)")?\)'): ExtractType.TEXT,
|
|
re.compile(r'TTRC\("(?P<message>([^"\\]|\\.)*)"\)'): ExtractType.TEXT,
|
|
re.compile(
|
|
r'TTRN\("(?P<message>([^"\\]|\\.)*)", "(?P<plural_message>([^"\\]|\\.)*)",[^,)]+?(, "(?P<context>([^"\\]|\\.)*)")?\)'
|
|
): ExtractType.TEXT,
|
|
re.compile(
|
|
r'RTRN\("(?P<message>([^"\\]|\\.)*)", "(?P<plural_message>([^"\\]|\\.)*)",[^,)]+?(, "(?P<context>([^"\\]|\\.)*)")?\)'
|
|
): ExtractType.TEXT,
|
|
re.compile(r'_initial_set\("(?P<message>[^"]+?)",'): ExtractType.PROPERTY_PATH,
|
|
re.compile(r'GLOBAL_DEF(_RST)?(_NOVAL)?\("(?P<message>[^".]+?)",'): ExtractType.PROPERTY_PATH,
|
|
re.compile(r'EDITOR_DEF(_RST)?\("(?P<message>[^"]+?)",'): ExtractType.PROPERTY_PATH,
|
|
re.compile(
|
|
r'(ADD_PROPERTYI?|ImportOption|ExportOption)\(PropertyInfo\(Variant::[_A-Z0-9]+, "(?P<message>[^"]+?)"[,)]'
|
|
): ExtractType.PROPERTY_PATH,
|
|
re.compile(
|
|
r"(?!#define )LIMPL_PROPERTY(_RANGE)?\(Variant::[_A-Z0-9]+, (?P<message>[^,]+?),"
|
|
): ExtractType.PROPERTY_PATH,
|
|
re.compile(r'ADD_GROUP\("(?P<message>[^"]+?)", "(?P<prefix>[^"]*?)"\)'): ExtractType.GROUP,
|
|
}
|
|
|
|
|
|
# See String::camelcase_to_underscore().
|
|
capitalize_re = re.compile(r"(?<=\D)(?=\d)|(?<=\d)(?=\D([a-z]|\d))")
|
|
|
|
|
|
def _process_editor_string(name):
|
|
# See EditorPropertyNameProcessor::process_string().
|
|
capitalized_parts = []
|
|
for segment in name.split("_"):
|
|
if not segment:
|
|
continue
|
|
remapped = remaps.get(segment)
|
|
if remapped:
|
|
capitalized_parts.append(remapped)
|
|
else:
|
|
# See String::capitalize().
|
|
# fmt: off
|
|
capitalized_parts.append(" ".join(
|
|
part.title()
|
|
for part in capitalize_re.sub("_", segment).replace("_", " ").split()
|
|
))
|
|
# fmt: on
|
|
|
|
return " ".join(capitalized_parts)
|
|
|
|
|
|
def _is_block_translator_comment(translator_line):
|
|
line = translator_line.strip()
|
|
if line.find("//") == 0:
|
|
return False
|
|
else:
|
|
return True
|
|
|
|
|
|
def _extract_translator_comment(line, is_block_translator_comment):
|
|
line = line.strip()
|
|
reached_end = False
|
|
extracted_comment = ""
|
|
|
|
start = line.find("TRANSLATORS:")
|
|
if start == -1:
|
|
start = 0
|
|
else:
|
|
start += len("TRANSLATORS:")
|
|
|
|
if is_block_translator_comment:
|
|
# If '*/' is found, then it's the end.
|
|
if line.rfind("*/") != -1:
|
|
extracted_comment = line[start : line.rfind("*/")]
|
|
reached_end = True
|
|
else:
|
|
extracted_comment = line[start:]
|
|
else:
|
|
# If beginning is not '//', then it's the end.
|
|
if line.find("//") != 0:
|
|
reached_end = True
|
|
else:
|
|
start = 2 if start == 0 else start
|
|
extracted_comment = line[start:]
|
|
|
|
return (not reached_end, extracted_comment)
|
|
|
|
|
|
def process_file(f, fname):
|
|
l = f.readline()
|
|
lc = 1
|
|
reading_translator_comment = False
|
|
is_block_translator_comment = False
|
|
translator_comment = ""
|
|
current_group = ""
|
|
|
|
while l:
|
|
|
|
# Detect translator comments.
|
|
if not reading_translator_comment and l.find("TRANSLATORS:") != -1:
|
|
reading_translator_comment = True
|
|
is_block_translator_comment = _is_block_translator_comment(l)
|
|
translator_comment = ""
|
|
|
|
# Gather translator comments. It will be gathered for the next translation function.
|
|
if reading_translator_comment:
|
|
reading_translator_comment, extracted_comment = _extract_translator_comment(l, is_block_translator_comment)
|
|
if extracted_comment != "":
|
|
translator_comment += extracted_comment + "\n"
|
|
if not reading_translator_comment:
|
|
translator_comment = translator_comment[:-1] # Remove extra \n at the end.
|
|
|
|
if not reading_translator_comment:
|
|
for pattern, extract_type in message_patterns.items():
|
|
for m in pattern.finditer(l):
|
|
location = os.path.relpath(fname).replace("\\", "/")
|
|
if line_nb:
|
|
location += ":" + str(lc)
|
|
|
|
captures = m.groupdict("")
|
|
msg = captures.get("message", "")
|
|
msg_plural = captures.get("plural_message", "")
|
|
msgctx = captures.get("context", "")
|
|
|
|
if extract_type == ExtractType.TEXT:
|
|
_add_message(msg, msg_plural, msgctx, location, translator_comment)
|
|
elif extract_type == ExtractType.PROPERTY_PATH:
|
|
if current_group:
|
|
if msg.startswith(current_group):
|
|
msg = msg[len(current_group) :]
|
|
else:
|
|
current_group = ""
|
|
for part in msg.split("/"):
|
|
_add_message(_process_editor_string(part), msg_plural, msgctx, location, translator_comment)
|
|
elif extract_type == ExtractType.GROUP:
|
|
_add_message(msg, msg_plural, msgctx, location, translator_comment)
|
|
current_group = captures["prefix"]
|
|
translator_comment = ""
|
|
|
|
l = f.readline()
|
|
lc += 1
|
|
|
|
|
|
def _add_message(msg, msg_plural, msgctx, location, translator_comment):
|
|
key = (msg, msgctx)
|
|
message = messages_map.get(key)
|
|
if not message:
|
|
message = Message()
|
|
message.msgid = msg
|
|
message.msgid_plural = msg_plural
|
|
message.msgctxt = msgctx
|
|
message.locations = []
|
|
message.comments = []
|
|
messages_map[key] = message
|
|
if location not in message.locations:
|
|
message.locations.append(location)
|
|
if translator_comment and translator_comment not in message.comments:
|
|
message.comments.append(translator_comment)
|
|
|
|
|
|
print("Updating the editor.pot template...")
|
|
|
|
for fname in matches:
|
|
with open(fname, "r", encoding="utf8") as f:
|
|
process_file(f, fname)
|
|
|
|
main_po += "\n\n".join(message.format() for message in messages_map.values())
|
|
|
|
with open("editor.pot", "w") as f:
|
|
f.write(main_po)
|
|
|
|
if os.name == "posix":
|
|
print("Wrapping template at 79 characters for compatibility with Weblate.")
|
|
os.system("msgmerge -w79 editor.pot editor.pot > editor.pot.wrap")
|
|
shutil.move("editor.pot.wrap", "editor.pot")
|
|
|
|
shutil.move("editor.pot", "editor/translations/editor.pot")
|
|
|
|
# TODO: Make that in a portable way, if we care; if not, kudos to Unix users
|
|
if os.name == "posix":
|
|
added = subprocess.check_output(r"git diff editor/translations/editor.pot | grep \+msgid | wc -l", shell=True)
|
|
removed = subprocess.check_output(r"git diff editor/translations/editor.pot | grep \\\-msgid | wc -l", shell=True)
|
|
print("\n# Template changes compared to the staged status:")
|
|
print("# Additions: %s msgids.\n# Deletions: %s msgids." % (int(added), int(removed)))
|