v8/tools/generate-header-include-checks.py
Vadim Gorbachev (bmsdave) 7315d7b3d7 Preparing v8 to use with python3 /tools
There are now less that 400 days until the end of life
of Python 2(aka _legacy_ Python) https://pythonclock.org/ .
The code compatibility check for python2 and python3
used the following tools: futurize, flake8
You can see the reports here: https://travis-ci.com/bmsdave/v8/builds

This CL was uploaded by git cl split.

Bug: v8:8594
Change-Id: I661c52a70527e8ddde841fee6d4dcba282b4a938
Reviewed-on: https://chromium-review.googlesource.com/c/1470123
Commit-Queue: Sergiy Belozorov <sergiyb@chromium.org>
Reviewed-by: Sergiy Belozorov <sergiyb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#59675}
2019-02-19 09:12:07 +00:00

147 lines
4.8 KiB
Python
Executable File

#!/usr/bin/env python
# vim:fenc=utf-8:shiftwidth=2
# Copyright 2018 the V8 project authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
"""Check that each header can be included in isolation.
For each header we generate one .cc file which only includes this one header.
All these .cc files are then added to a sources.gni file which is included in
BUILD.gn. Just compile to check whether there are any violations to the rule
that each header must be includable in isolation.
"""
# for py2/py3 compatibility
from __future__ import print_function
import argparse
import os
import os.path
import re
import sys
# TODO(clemensh): Extend to tests.
DEFAULT_INPUT = ['base', 'src']
DEFAULT_GN_FILE = 'BUILD.gn'
MY_DIR = os.path.dirname(os.path.realpath(__file__))
V8_DIR = os.path.dirname(MY_DIR)
OUT_DIR = os.path.join(V8_DIR, 'check-header-includes')
AUTO_EXCLUDE = [
# flag-definitions.h needs a mode set for being included.
'src/flag-definitions.h',
]
AUTO_EXCLUDE_PATTERNS = [
'src/base/atomicops_internals_.*',
] + [
# platform-specific headers
'\\b{}\\b'.format(p) for p in
('win', 'win32', 'ia32', 'x64', 'arm', 'arm64', 'mips', 'mips64', 's390',
'ppc')]
args = None
def parse_args():
global args
parser = argparse.ArgumentParser()
parser.add_argument('-i', '--input', type=str, action='append',
help='Headers or directories to check (directories '
'are scanned for headers recursively); default: ' +
','.join(DEFAULT_INPUT))
parser.add_argument('-x', '--exclude', type=str, action='append',
help='Add an exclude pattern (regex)')
parser.add_argument('-v', '--verbose', action='store_true',
help='Be verbose')
args = parser.parse_args()
args.exclude = (args.exclude or []) + AUTO_EXCLUDE_PATTERNS
args.exclude += ['^' + re.escape(x) + '$' for x in AUTO_EXCLUDE]
if not args.input:
args.input=DEFAULT_INPUT
def printv(line):
if args.verbose:
print(line)
def find_all_headers():
printv('Searching for headers...')
header_files = []
exclude_patterns = [re.compile(x) for x in args.exclude]
def add_recursively(filename):
full_name = os.path.join(V8_DIR, filename)
if not os.path.exists(full_name):
sys.exit('File does not exist: {}'.format(full_name))
if os.path.isdir(full_name):
for subfile in os.listdir(full_name):
full_name = os.path.join(filename, subfile)
printv('Scanning {}'.format(full_name))
add_recursively(full_name)
elif filename.endswith('.h'):
printv('--> Found header file {}'.format(filename))
for p in exclude_patterns:
if p.search(filename):
printv('--> EXCLUDED (matches {})'.format(p.pattern))
return
header_files.append(filename)
for filename in args.input:
add_recursively(filename)
return header_files
def get_cc_file_name(header):
split = os.path.split(header)
header_dir = os.path.relpath(split[0], V8_DIR)
# Prefix with the directory name, to avoid collisions in the object files.
prefix = header_dir.replace(os.path.sep, '-')
cc_file_name = 'test-include-' + prefix + '-' + split[1][:-1] + 'cc'
return os.path.join(OUT_DIR, cc_file_name)
def create_including_cc_files(header_files):
comment = 'check including this header in isolation'
for header in header_files:
cc_file_name = get_cc_file_name(header)
rel_cc_file_name = os.path.relpath(cc_file_name, V8_DIR)
content = '#include "{}" // {}\n'.format(header, comment)
if os.path.exists(cc_file_name):
with open(cc_file_name) as cc_file:
if cc_file.read() == content:
printv('File {} is up to date'.format(rel_cc_file_name))
continue
printv('Creating file {}'.format(rel_cc_file_name))
with open(cc_file_name, 'w') as cc_file:
cc_file.write(content)
def generate_gni(header_files):
gni_file = os.path.join(OUT_DIR, 'sources.gni')
printv('Generating file "{}"'.format(os.path.relpath(gni_file, V8_DIR)))
with open(gni_file, 'w') as gn:
gn.write("""\
# Copyright 2018 The Chromium Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
# This list is filled automatically by tools/check_header_includes.py.
check_header_includes_sources = [
""");
for header in header_files:
cc_file_name = get_cc_file_name(header)
gn.write(' "{}",\n'.format(os.path.relpath(cc_file_name, V8_DIR)))
gn.write(']\n')
def main():
parse_args()
header_files = find_all_headers()
if not os.path.exists(OUT_DIR):
os.mkdir(OUT_DIR)
create_including_cc_files(header_files)
generate_gni(header_files)
if __name__ == '__main__':
main()