You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

425 lines
15 KiB

# Copyright (C) 2018 and later: Unicode, Inc. and others.
# License & terms of use: http://www.unicode.org/copyright.html
# Python 2/3 Compatibility (ICU-20299)
# TODO(ICU-20301): Remove this.
from __future__ import print_function
from abc import abstractmethod
from collections import defaultdict
import re
import sys
from . import *
from . import utils
from .request_types import *
# Note: for this to be a proper abstract class, it should extend abc.ABC.
# There is no nice way to do this that works in both Python 2 and 3.
# TODO(ICU-20301): Make this inherit from abc.ABC.
class Filter(object):
@staticmethod
def create_from_json(json_data, io):
assert io != None
if "filterType" in json_data:
filter_type = json_data["filterType"]
else:
filter_type = "file-stem"
if filter_type == "file-stem":
return FileStemFilter(json_data)
elif filter_type == "language":
return LanguageFilter(json_data)
elif filter_type == "regex":
return RegexFilter(json_data)
elif filter_type == "exclude":
return ExclusionFilter()
elif filter_type == "union":
return UnionFilter(json_data, io)
elif filter_type == "locale":
return LocaleFilter(json_data, io)
else:
print("Error: Unknown filterType option: %s" % filter_type, file=sys.stderr)
return None
def filter(self, request):
if not request.apply_file_filter(self):
return []
for file in request.all_input_files():
assert self.match(file)
return [request]
@staticmethod
def _file_to_file_stem(file):
start = file.filename.rfind("/")
limit = file.filename.rfind(".")
return file.filename[start+1:limit]
@staticmethod
def _file_to_subdir(file):
limit = file.filename.rfind("/")
if limit == -1:
return None
return file.filename[:limit]
@abstractmethod
def match(self, file):
pass
class InclusionFilter(Filter):
def match(self, file):
return True
class ExclusionFilter(Filter):
def match(self, file):
return False
class IncludeExcludeFilter(Filter):
def __init__(self, json_data):
if "whitelist" in json_data:
self.is_includelist = True
self.includelist = json_data["whitelist"]
elif "includelist" in json_data:
self.is_includelist = True
self.includelist = json_data["includelist"]
elif "blacklist" in json_data:
self.is_includelist = False
self.excludelist = json_data["blacklist"]
elif "excludelist" in json_data:
self.is_includelist = False
self.excludelist = json_data["excludelist"]
else:
raise AssertionError("Need either includelist or excludelist: %s" % str(json_data))
def match(self, file):
file_stem = self._file_to_file_stem(file)
return self._should_include(file_stem)
@abstractmethod
def _should_include(self, file_stem):
pass
class FileStemFilter(IncludeExcludeFilter):
def _should_include(self, file_stem):
if self.is_includelist:
return file_stem in self.includelist
else:
return file_stem not in self.excludelist
class LanguageFilter(IncludeExcludeFilter):
def _should_include(self, file_stem):
language = file_stem.split("_")[0]
if language == "root":
# Always include root.txt
return True
if self.is_includelist:
return language in self.includelist
else:
return language not in self.excludelist
class RegexFilter(IncludeExcludeFilter):
def __init__(self, *args):
# TODO(ICU-20301): Change this to: super().__init__(*args)
super(RegexFilter, self).__init__(*args)
if self.is_includelist:
self.includelist = [re.compile(pat) for pat in self.includelist]
else:
self.excludelist = [re.compile(pat) for pat in self.excludelist]
def _should_include(self, file_stem):
if self.is_includelist:
for pattern in self.includelist:
if pattern.match(file_stem):
return True
return False
else:
for pattern in self.excludelist:
if pattern.match(file_stem):
return False
return True
class UnionFilter(Filter):
def __init__(self, json_data, io):
# Collect the sub-filters.
self.sub_filters = []
for filter_json in json_data["unionOf"]:
self.sub_filters.append(Filter.create_from_json(filter_json, io))
def match(self, file):
"""Match iff any of the sub-filters match."""
for filter in self.sub_filters:
if filter.match(file):
return True
return False
LANGUAGE_SCRIPT_REGEX = re.compile(r"^([a-z]{2,3})_[A-Z][a-z]{3}$")
LANGUAGE_ONLY_REGEX = re.compile(r"^[a-z]{2,3}$")
class LocaleFilter(Filter):
def __init__(self, json_data, io):
if "whitelist" in json_data:
self.locales_requested = list(json_data["whitelist"])
elif "includelist" in json_data:
self.locales_requested = list(json_data["includelist"])
else:
raise AssertionError("You must have an includelist in a locale filter")
self.include_children = json_data.get("includeChildren", True)
self.include_scripts = json_data.get("includeScripts", False)
# Load the dependency graph from disk
self.dependency_data_by_tree = {
tree: io.read_locale_deps(tree)
for tree in utils.ALL_TREES
}
def match(self, file):
tree = self._file_to_subdir(file)
assert tree is not None
locale = self._file_to_file_stem(file)
# A locale is *required* if it is *requested* or an ancestor of a
# *requested* locale.
if locale in self._locales_required(tree):
return True
# Resolve include_scripts and include_children.
return self._match_recursive(locale, tree)
def _match_recursive(self, locale, tree):
# Base case: return True if we reached a *requested* locale,
# or False if we ascend out of the locale tree.
if locale is None:
return False
if locale in self.locales_requested:
return True
# Check for alternative scripts.
# This causes sr_Latn to check sr instead of going directly to root.
if self.include_scripts:
match = LANGUAGE_SCRIPT_REGEX.match(locale)
if match and self._match_recursive(match.group(1), tree):
return True
# Check if we are a descendant of a *requested* locale.
if self.include_children:
parent = self._get_parent_locale(locale, tree)
if self._match_recursive(parent, tree):
return True
# No matches.
return False
def _get_parent_locale(self, locale, tree):
"""Gets the parent locale in the given tree, according to dependency data."""
dependency_data = self.dependency_data_by_tree[tree]
if "parents" in dependency_data and locale in dependency_data["parents"]:
return dependency_data["parents"][locale]
if "aliases" in dependency_data and locale in dependency_data["aliases"]:
return dependency_data["aliases"][locale]
if LANGUAGE_ONLY_REGEX.match(locale):
return "root"
i = locale.rfind("_")
if i < 0:
assert locale == "root", "Invalid locale: %s/%s" % (tree, locale)
return None
return locale[:i]
def _locales_required(self, tree):
"""Returns a generator of all required locales in the given tree."""
for locale in self.locales_requested:
while locale is not None:
yield locale
locale = self._get_parent_locale(locale, tree)
def apply_filters(requests, config, io):
"""Runs the filters and returns a new list of requests."""
requests = _apply_file_filters(requests, config, io)
requests = _apply_resource_filters(requests, config, io)
return requests
def _apply_file_filters(old_requests, config, io):
"""Filters out entire files."""
filters = _preprocess_file_filters(old_requests, config, io)
new_requests = []
for request in old_requests:
category = request.category
if category in filters:
new_requests += filters[category].filter(request)
else:
new_requests.append(request)
return new_requests
def _preprocess_file_filters(requests, config, io):
all_categories = set(
request.category
for request in requests
)
all_categories.remove(None)
all_categories = list(sorted(all_categories))
json_data = config.filters_json_data
filters = {}
default_filter_json = "exclude" if config.strategy == "additive" else "include"
for category in all_categories:
filter_json = default_filter_json
# Figure out the correct filter to create
if "featureFilters" in json_data and category in json_data["featureFilters"]:
filter_json = json_data["featureFilters"][category]
if filter_json == "include" and "localeFilter" in json_data and category.endswith("_tree"):
filter_json = json_data["localeFilter"]
# Resolve the filter JSON into a filter object
if filter_json == "exclude":
filters[category] = ExclusionFilter()
elif filter_json == "include":
pass # no-op
else:
filters[category] = Filter.create_from_json(filter_json, io)
if "featureFilters" in json_data:
for category in json_data["featureFilters"]:
if category not in all_categories:
print("Warning: category %s is not known" % category, file=sys.stderr)
return filters
class ResourceFilterInfo(object):
def __init__(self, category, strategy):
self.category = category
self.strategy = strategy
self.filter_tmp_dir = "filters/%s" % category
self.input_files = None
self.filter_files = None
self.rules_by_file = None
def apply_to_requests(self, all_requests):
# Call this method only once per list of requests.
assert self.input_files is None
for request in all_requests:
if request.category != self.category:
continue
if not isinstance(request, AbstractExecutionRequest):
continue
if request.tool != IcuTool("genrb"):
continue
if not request.input_files:
continue
self._set_files(request.input_files)
request.dep_targets += [self.filter_files[:]]
arg_str = "--filterDir {TMP_DIR}/%s" % self.filter_tmp_dir
request.args = "%s %s" % (arg_str, request.args)
# Make sure we found the target request
if self.input_files is None:
print("WARNING: Category not found: %s" % self.category, file=sys.stderr)
self.input_files = []
self.filter_files = []
self.rules_by_file = []
def _set_files(self, files):
# Note: The input files to genrb for a certain category should always
# be the same. For example, there are often two genrb calls: one for
# --writePoolBundle, and the other for --usePoolBundle. They are both
# expected to have the same list of input files.
if self.input_files is not None:
assert self.input_files == files
return
self.input_files = list(files)
self.filter_files = [
TmpFile("%s/%s" % (self.filter_tmp_dir, basename))
for basename in (
file.filename[file.filename.rfind("/")+1:]
for file in files
)
]
if self.strategy == "additive":
self.rules_by_file = [
[r"-/", r"+/%%ALIAS", r"+/%%Parent"]
for _ in range(len(files))
]
else:
self.rules_by_file = [
[r"+/"]
for _ in range(len(files))
]
def add_rules(self, file_filter, rules):
for file, rule_list in zip(self.input_files, self.rules_by_file):
if file_filter.match(file):
rule_list += rules
def make_requests(self):
# Map from rule list to filter files with that rule list
unique_rules = defaultdict(list)
for filter_file, rules in zip(self.filter_files, self.rules_by_file):
unique_rules[tuple(rules)].append(filter_file)
new_requests = []
i = 0
for rules, filter_files in unique_rules.items():
base_filter_file = filter_files[0]
new_requests += [
PrintFileRequest(
name = "%s_print_%d" % (self.category, i),
output_file = base_filter_file,
content = self._generate_resource_filter_txt(rules)
)
]
i += 1
for filter_file in filter_files[1:]:
new_requests += [
CopyRequest(
name = "%s_copy_%d" % (self.category, i),
input_file = base_filter_file,
output_file = filter_file
)
]
i += 1
return new_requests
@staticmethod
def _generate_resource_filter_txt(rules):
result = "# Caution: This file is automatically generated\n\n"
result += "\n".join(rules)
return result
def _apply_resource_filters(all_requests, config, io):
"""Creates filters for looking within resource bundle files."""
json_data = config.filters_json_data
if "resourceFilters" not in json_data:
return all_requests
collected = {}
for entry in json_data["resourceFilters"]:
if "files" in entry:
file_filter = Filter.create_from_json(entry["files"], io)
else:
file_filter = InclusionFilter()
for category in entry["categories"]:
# not defaultdict because we need to pass arguments to the constructor
if category not in collected:
filter_info = ResourceFilterInfo(category, config.strategy)
filter_info.apply_to_requests(all_requests)
collected[category] = filter_info
else:
filter_info = collected[category]
filter_info.add_rules(file_filter, entry["rules"])
# Add the filter generation requests to the beginning so that by default
# they are made before genrb gets run (order is required by windirect)
new_requests = []
for filter_info in collected.values():
new_requests += filter_info.make_requests()
new_requests += all_requests
return new_requests