466 lines
20 KiB
Python
466 lines
20 KiB
Python
###########################################################################
|
|
# This file provides maintenance on the various language files
|
|
# 1. Create new "xx/cards_xx.json" files that have entries ordered as:
|
|
# a. the card_tag entries in "cards_db.json"
|
|
# b. the group_tag entries as found in "cards_db.json"
|
|
# c. the super group entries (grouping across all expansions"
|
|
# d. any unused entries existing in the file (assumed to be work in progress)
|
|
#
|
|
# 2. Create new "sets_db.json" and "xx/cards_xx.json" with entries sorted alphabetically
|
|
#
|
|
# All output is in the designated output directory. Original files are not overwritten.
|
|
###########################################################################
|
|
from __future__ import print_function
|
|
import os
|
|
import os.path
|
|
import io
|
|
import codecs
|
|
import json
|
|
from shutil import copyfile
|
|
import argparse
|
|
|
|
import six
|
|
|
|
LANGUAGE_DEFAULT = 'en_us' # default language, which takes priority
|
|
LANGUAGE_XX = 'xx' # language for starting a translation
|
|
|
|
|
|
def get_lang_dirs(path):
|
|
# Find all valid languages.
|
|
languages = []
|
|
for name in os.listdir(path):
|
|
dir_path = os.path.join(path, name)
|
|
if os.path.isdir(dir_path):
|
|
cards_file = os.path.join(dir_path, "cards_" + name + ".json")
|
|
sets_file = os.path.join(dir_path, "sets_" + name + ".json")
|
|
if os.path.isfile(cards_file) and os.path.isfile(sets_file):
|
|
languages.append(name)
|
|
return languages
|
|
|
|
|
|
def get_json_data(json_file_path):
|
|
print(('reading {}'.format(json_file_path)))
|
|
# Read in the json from the specified file
|
|
with codecs.open(json_file_path, 'r', 'utf-8') as json_file:
|
|
data = json.load(json_file)
|
|
assert data, "Could not load json at: '%r' " % json_file_path
|
|
return data
|
|
|
|
|
|
def json_dict_entry(entry, separator=''):
|
|
# Return a nicely formated json dict entry.
|
|
# It does not include the enclosing {} and removes trailing white space
|
|
json_data = json.dumps(entry, indent=4, ensure_ascii=False, sort_keys=True)
|
|
json_data = json_data.strip(
|
|
'{}').rstrip() # Remove outer{} and then trailing whitespace
|
|
return separator + json_data
|
|
|
|
|
|
# Multikey sort
|
|
# see: http://stackoverflow.com/questions/1143671/python-sorting-list-of-dictionaries-by-multiple-keys
|
|
def multikeysort(items, columns):
|
|
from operator import itemgetter
|
|
for c in columns[::-1]:
|
|
items = sorted(items, key=itemgetter(c))
|
|
return items
|
|
|
|
|
|
def main(args):
|
|
###########################################################################
|
|
# Get all the languages, and place the default language first in the list
|
|
###########################################################################
|
|
languages = get_lang_dirs(args.card_db_dir)
|
|
languages.remove(LANGUAGE_DEFAULT)
|
|
languages.insert(0, LANGUAGE_DEFAULT)
|
|
if LANGUAGE_XX not in languages:
|
|
languages.append(LANGUAGE_XX)
|
|
print("Languages:")
|
|
print(languages)
|
|
print()
|
|
|
|
###########################################################################
|
|
# Make sure the directories exist to hold the output
|
|
###########################################################################
|
|
|
|
# main output directory
|
|
if not os.path.exists(args.output_dir):
|
|
os.makedirs(args.output_dir)
|
|
|
|
# each language directory
|
|
for lang in languages:
|
|
# Make sure the directory is there to hold the file
|
|
lang_dir = os.path.join(args.output_dir, lang)
|
|
if not os.path.exists(lang_dir):
|
|
os.makedirs(lang_dir)
|
|
|
|
###########################################################################
|
|
# Get the types_db information
|
|
# Store in a list in the order found in types[]. Ordered by card_type
|
|
# 1. card_tags, 2. group_tags, 3. super groups
|
|
###########################################################################
|
|
type_parts = []
|
|
|
|
# Get the card data
|
|
type_data = get_json_data(os.path.join(args.card_db_dir, "types_db.json"))
|
|
|
|
# Sort the cards by cardset_tags, then card_tag
|
|
sorted_type_data = multikeysort(type_data, ['card_type'])
|
|
|
|
with io.open(
|
|
os.path.join(args.output_dir, "types_db.json"), 'w',
|
|
encoding='utf-8') as lang_out:
|
|
lang_out.write(six.u("[")) # Start of list
|
|
sep = ""
|
|
for type in sorted_type_data:
|
|
# Collect all the individual types
|
|
type_parts = list(set(type['card_type']) | set(type_parts))
|
|
lang_out.write(sep + json.dumps(
|
|
type, indent=4, ensure_ascii=False, sort_keys=True))
|
|
sep = ","
|
|
lang_out.write(six.u("\n]\n")) # End of List
|
|
|
|
type_parts.sort()
|
|
print("Unique Types:")
|
|
print(type_parts)
|
|
print()
|
|
|
|
###########################################################################
|
|
# Get the labels_db information
|
|
# Store in a list in the order found.
|
|
###########################################################################
|
|
all_labels = []
|
|
|
|
# Get the card data
|
|
label_data = get_json_data(os.path.join(args.card_db_dir, "labels_db.json"))
|
|
|
|
with io.open(
|
|
os.path.join(args.output_dir, "labels_db.json"), 'w',
|
|
encoding='utf-8') as lang_out:
|
|
lang_out.write(six.u("[")) # Start of list
|
|
sep = ""
|
|
for label in label_data:
|
|
# Collect all the individual types
|
|
all_labels = list(set(label['names']) | set(all_labels))
|
|
lang_out.write(sep + json.dumps(
|
|
label, indent=4, ensure_ascii=False, sort_keys=True))
|
|
sep = ","
|
|
lang_out.write(six.u("\n]\n")) # End of List
|
|
|
|
all_labels.sort()
|
|
print("Labels: ")
|
|
print(all_labels)
|
|
print()
|
|
###########################################################################
|
|
# Fix up all the xx/types_xx.json files
|
|
# Place entries in alphabetical order
|
|
# If entries don't exist:
|
|
# If the default language, set from information in the "types_db.json" file,
|
|
# If not the default language, set based on information from the default language.
|
|
# Lastly, keep any extra entries that are not currently used, just in case needed
|
|
# in the future or is a work in progress.
|
|
###########################################################################
|
|
for lang in languages:
|
|
lang_file = "types_" + lang + ".json"
|
|
fname = os.path.join(args.card_db_dir, lang, lang_file)
|
|
if os.path.isfile(fname):
|
|
lang_type_data = get_json_data(fname)
|
|
else:
|
|
lang_type_data = {}
|
|
|
|
with io.open(
|
|
os.path.join(args.output_dir, lang, lang_file), 'w',
|
|
encoding='utf-8') as lang_out:
|
|
lang_out.write(six.u("{")) # Start of types
|
|
sep = ""
|
|
used = []
|
|
|
|
for type in sorted(type_parts):
|
|
if type not in lang_type_data:
|
|
if lang == LANGUAGE_DEFAULT:
|
|
lang_type_data[type] = type
|
|
lang_type_default = lang_type_data
|
|
else:
|
|
lang_type_data[type] = lang_type_default[type]
|
|
|
|
lang_out.write(json_dict_entry({type: lang_type_data[type]}, sep))
|
|
used.append(type)
|
|
sep = ","
|
|
|
|
# Now keep any unused values just in case needed in the future
|
|
for key in lang_type_data:
|
|
if key not in used:
|
|
lang_out.write(
|
|
json_dict_entry({
|
|
key: lang_type_data[key]
|
|
}, sep))
|
|
sep = ","
|
|
|
|
lang_out.write(six.u("\n}\n")) # End of Types
|
|
|
|
if lang == LANGUAGE_DEFAULT:
|
|
lang_type_default = lang_type_data # Keep for later languages
|
|
|
|
###########################################################################
|
|
# Get the cards_db information
|
|
# Store in a list in the order found in cards[]. Ordered as follows:
|
|
# 1. card_tags, 2. group_tags, 3. super groups
|
|
###########################################################################
|
|
cards = []
|
|
groups = []
|
|
super_groups = [u'events', u'landmarks']
|
|
|
|
# Get the card data
|
|
card_data = get_json_data(os.path.join(args.card_db_dir, "cards_db.json"))
|
|
|
|
# Sort the cardset_tags
|
|
for card in card_data:
|
|
card["cardset_tags"].sort()
|
|
# But put all the base cards together by moving to front of the list
|
|
if 'base' in card["cardset_tags"]:
|
|
card["cardset_tags"].remove('base')
|
|
card["cardset_tags"].insert(0, 'base')
|
|
|
|
# Sort the cards by cardset_tags, then card_tag
|
|
sorted_card_data = multikeysort(card_data, ['cardset_tags', 'card_tag'])
|
|
|
|
with io.open(
|
|
os.path.join(args.output_dir, "cards_db.json"), 'w',
|
|
encoding='utf-8') as lang_out:
|
|
lang_out.write(six.u("[")) # Start of list
|
|
sep = ""
|
|
for card in sorted_card_data:
|
|
if card['card_tag'] not in cards:
|
|
cards.append(card['card_tag'])
|
|
if 'group_tag' in card:
|
|
if card['group_tag'] not in groups:
|
|
groups.append(card['group_tag'])
|
|
lang_out.write(sep + json.dumps(
|
|
card, indent=4, ensure_ascii=False, sort_keys=True))
|
|
sep = ","
|
|
lang_out.write(six.u("\n]\n")) # End of List
|
|
|
|
cards.extend(groups)
|
|
cards.extend(super_groups)
|
|
|
|
print("Cards:")
|
|
print(cards)
|
|
print()
|
|
|
|
###########################################################################
|
|
# Fix up all the cards_xx.json files
|
|
# Place entries in the same order as given in "cards_db.json".
|
|
# If entries don't exist:
|
|
# If the default language, set base on information in the "cards_db.json" file,
|
|
# If not the default language, set based on information from the default language.
|
|
# Lastly, keep any extra entries that are not currently used, just in case needed
|
|
# in the future or is a work in progress.
|
|
###########################################################################
|
|
for lang in languages:
|
|
|
|
# contruct the cards json file name
|
|
lang_file = "cards_" + lang + ".json"
|
|
fname = os.path.join(args.card_db_dir, lang, lang_file)
|
|
if os.path.isfile(fname):
|
|
lang_data = get_json_data(fname)
|
|
else:
|
|
lang_data = {}
|
|
|
|
# Process the file
|
|
with io.open(
|
|
os.path.join(args.output_dir, lang, lang_file), 'w',
|
|
encoding='utf-8') as lang_out:
|
|
lang_out.write(six.u("{")) # Start of set
|
|
sep = ""
|
|
fields = [u"description", u"extra", u"name"]
|
|
|
|
for card in cards:
|
|
if card not in lang_data or lang == LANGUAGE_XX:
|
|
# Card is missing, need to add it
|
|
lang_data[card] = {}
|
|
if lang == LANGUAGE_DEFAULT:
|
|
# Default language gets bare minimum. Really need to add by hand.
|
|
lang_data[card]["extra"] = ""
|
|
lang_data[card]["name"] = card
|
|
lang_data[card]["description"] = ""
|
|
lang_data[card]["untranslated"] = ', '.join(fields)
|
|
lang_default = lang_data
|
|
else:
|
|
# All other languages should get the default languages' text
|
|
lang_data[card]["extra"] = lang_default[card]["extra"]
|
|
lang_data[card]["name"] = lang_default[card]["name"]
|
|
lang_data[card]["description"] = lang_default[card][
|
|
"description"]
|
|
lang_data[card]["untranslated"] = ', '.join(fields)
|
|
else:
|
|
# Card exists, figure out what needs updating (don't update default language)
|
|
if lang != LANGUAGE_DEFAULT:
|
|
if 'untranslated' in lang_data[card]:
|
|
# Has an 'untranslated' field. Process accordingly
|
|
if not lang_data[card]["untranslated"].strip():
|
|
# It is empty, so just remove it
|
|
del lang_data[card]["untranslated"]
|
|
else:
|
|
# If a field remains untranslated, then replace with the default languages copy
|
|
for field in fields:
|
|
if field in lang_data[card]['untranslated']:
|
|
lang_data[card][field] = lang_default[
|
|
card][field]
|
|
else:
|
|
# Need to create the 'untranslated' field and update based upon existing fields
|
|
untranslated = []
|
|
for field in fields:
|
|
if field not in lang_data[card]:
|
|
lang_data[card][field] = lang_default[card][
|
|
field]
|
|
untranslated.append(field)
|
|
if untranslated:
|
|
# only add if something is still needing translation
|
|
lang_data[card]["untranslated"] = ', '.join(
|
|
untranslated)
|
|
|
|
lang_out.write(json_dict_entry({card: lang_data[card]}, sep))
|
|
lang_data[card]['used'] = True
|
|
sep = ","
|
|
|
|
# Now keep any unused values just in case needed in the future
|
|
for key in lang_data:
|
|
if 'used' not in lang_data[key]:
|
|
lang_data[key][
|
|
"untranslated"] = "Note: This card is currently not used."
|
|
lang_out.write(json_dict_entry({key: lang_data[key]}, sep))
|
|
sep = ","
|
|
lang_out.write(six.u("\n}\n")) # End of Set
|
|
|
|
if lang == LANGUAGE_DEFAULT:
|
|
lang_default = lang_data # Keep for later languages
|
|
|
|
###########################################################################
|
|
# Fix up the sets_db.json file
|
|
# Place entries in alphabetical order
|
|
###########################################################################
|
|
lang_file = "sets_db.json"
|
|
set_data = get_json_data(os.path.join(args.card_db_dir, lang_file))
|
|
|
|
with io.open(
|
|
os.path.join(args.output_dir, lang_file), 'w',
|
|
encoding='utf-8') as lang_out:
|
|
lang_out.write(six.u("{")) # Start of set
|
|
sep = ""
|
|
sets = []
|
|
for s in sorted(set_data):
|
|
lang_out.write(json_dict_entry({s: set_data[s]}, sep))
|
|
sep = ","
|
|
if s not in sets:
|
|
sets.append(s)
|
|
|
|
lang_out.write(six.u("\n}\n")) # End of Set
|
|
|
|
print("Sets:")
|
|
print(sets)
|
|
print()
|
|
|
|
###########################################################################
|
|
# Fix up all the xx/sets_xx.json files
|
|
# Place entries in alphabetical order
|
|
# If entries don't exist:
|
|
# If the default language, set from information in the "sets_db.json" file,
|
|
# If not the default language, set based on information from the default language.
|
|
# Lastly, keep any extra entries that are not currently used, just in case needed
|
|
# in the future or is a work in progress.
|
|
###########################################################################
|
|
for lang in languages:
|
|
lang_file = "sets_" + lang + ".json"
|
|
fname = os.path.join(args.card_db_dir, lang, lang_file)
|
|
if os.path.isfile(fname):
|
|
lang_set_data = get_json_data(fname)
|
|
else:
|
|
lang_set_data = {}
|
|
with io.open(
|
|
os.path.join(args.output_dir, lang, lang_file), 'w',
|
|
encoding='utf-8') as lang_out:
|
|
lang_out.write(six.u("{")) # Start of set
|
|
sep = ""
|
|
|
|
for s in sorted(set_data):
|
|
if s not in lang_set_data:
|
|
lang_set_data[s] = {}
|
|
if lang == LANGUAGE_DEFAULT:
|
|
lang_set_data[s]["set_name"] = s.title()
|
|
lang_set_data[s]["text_icon"] = set_data[s][
|
|
"text_icon"]
|
|
if 'short_name' in set_data[s]:
|
|
lang_set_data[s]["short_name"] = set_data[s][
|
|
"short_name"]
|
|
if 'set_text' in set_data[s]:
|
|
lang_set_data[s]["set_text"] = set_data[s][
|
|
"set_text"]
|
|
else:
|
|
lang_set_data[s]["set_name"] = lang_default[s][
|
|
"set_name"]
|
|
lang_set_data[s]["text_icon"] = lang_default[s][
|
|
"text_icon"]
|
|
if 'short_name' in lang_default[s]:
|
|
lang_set_data[s]["short_name"] = lang_default[s][
|
|
"short_name"]
|
|
if 'set_text' in lang_default[s]:
|
|
lang_set_data[s]["set_text"] = lang_default[s][
|
|
"set_text"]
|
|
else:
|
|
if lang != LANGUAGE_DEFAULT:
|
|
for x in lang_default[s]:
|
|
if x not in lang_set_data[s] and x is not 'used':
|
|
lang_set_data[s][x] = lang_default[s][x]
|
|
|
|
lang_out.write(json_dict_entry({s: lang_set_data[s]}, sep))
|
|
lang_set_data[s]['used'] = True
|
|
sep = ","
|
|
|
|
# Now keep any unused values just in case needed in the future
|
|
for key in lang_set_data:
|
|
if 'used' not in lang_set_data[key]:
|
|
lang_out.write(json_dict_entry({key: lang_set_data[key]}, sep))
|
|
sep = ","
|
|
|
|
lang_out.write(six.u("\n}\n")) # End of Set
|
|
|
|
if lang == LANGUAGE_DEFAULT:
|
|
lang_default = lang_set_data # Keep for later languages
|
|
|
|
###########################################################################
|
|
# bonuses_xx files
|
|
###########################################################################
|
|
for lang in languages:
|
|
# Special case for xx. Reseed from default language
|
|
fromLanguage = lang
|
|
if lang == LANGUAGE_XX:
|
|
fromLanguage = LANGUAGE_DEFAULT
|
|
|
|
copyfile(
|
|
os.path.join(args.card_db_dir, fromLanguage, "bonuses_" + fromLanguage + ".json"),
|
|
os.path.join(args.output_dir, lang, "bonuses_" + lang + ".json"))
|
|
|
|
###########################################################################
|
|
# translation.txt
|
|
###########################################################################
|
|
copyfile(
|
|
os.path.join(args.card_db_dir, "translation.md"),
|
|
os.path.join(args.output_dir, "translation.md"))
|
|
|
|
# Since xx is the starting point for new translations,
|
|
# make sure xx has the latest copy of translation.txt
|
|
copyfile(
|
|
os.path.join(args.card_db_dir, LANGUAGE_XX, "translation.txt"),
|
|
os.path.join(args.output_dir, LANGUAGE_XX, "translation.txt"))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument('--card_db_dir', default=os.path.join(os.path.dirname(os.path.abspath(__file__)),
|
|
"..", "domdiv", "card_db"),
|
|
help='directory of card data')
|
|
parser.add_argument('--output_dir',
|
|
default=os.path.join(os.path.dirname(os.path.abspath(__file__)), ".", "card_db"),
|
|
help='directory for output data')
|
|
args = parser.parse_args()
|
|
main(args)
|