Source code

Revision control

Copy as Markdown

Other Tools

# Copyright 2016 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.
from __future__ import print_function
import os
import struct
import sys
def Main(args):
if len(args) < 4:
print(
"Usage: %s output.hmap Foo.framework header1.h..." % args[0],
file=sys.stderr)
return 1
(out, framework, all_headers) = args[1], args[2], args[3:]
framework_name = os.path.basename(framework).split('.')[0]
all_headers = map(os.path.abspath, all_headers)
filelist = {}
for header in all_headers:
filename = os.path.basename(header)
filelist[filename] = header
filelist[os.path.join(framework_name, filename)] = header
WriteHmap(out, filelist)
return 0
def NextGreaterPowerOf2(x):
return 2**(x).bit_length()
def WriteHmap(output_name, filelist):
"""Generates a header map based on |filelist|.
Per Mark Mentovai:
A header map is structured essentially as a hash table, keyed by names used
in #includes, and providing pathnames to the actual files.
The implementation below and the comment above comes from inspecting:
while also looking at the implementation in clang in:
"""
magic = 1751998832
version = 1
_reserved = 0
count = len(filelist)
capacity = NextGreaterPowerOf2(count)
strings_offset = 24 + (12 * capacity)
max_value_length = len(max(filelist.values(), key=lambda v: len(v)))
out = open(output_name, 'wb')
out.write(struct.pack('<LHHLLLL', magic, version, _reserved, strings_offset,
count, capacity, max_value_length))
# Create empty hashmap buckets.
buckets = [None] * capacity
for file, path in filelist.items():
key = 0
for c in file:
key += ord(c.lower()) * 13
# Fill next empty bucket.
while buckets[key & capacity - 1] is not None:
key = key + 1
buckets[key & capacity - 1] = (file, path)
next_offset = 1
for bucket in buckets:
if bucket is None:
out.write(struct.pack('<LLL', 0, 0, 0))
else:
(file, path) = bucket
key_offset = next_offset
prefix_offset = key_offset + len(file) + 1
suffix_offset = prefix_offset + len(os.path.dirname(path) + os.sep) + 1
next_offset = suffix_offset + len(os.path.basename(path)) + 1
out.write(struct.pack('<LLL', key_offset, prefix_offset, suffix_offset))
# Pad byte since next offset starts at 1.
out.write(struct.pack('<x'))
for bucket in buckets:
if bucket is not None:
(file, path) = bucket
base = os.path.dirname(path) + os.sep
path = os.path.basename(path)
file = file.encode('UTF-8')
base = base.encode('UTF-8')
path = path.encode('UTF-8')
out.write(struct.pack('<%ds' % len(file), file))
out.write(struct.pack('<s', b'\0'))
out.write(struct.pack('<%ds' % len(base), base))
out.write(struct.pack('<s', b'\0'))
out.write(struct.pack('<%ds' % len(path), path))
out.write(struct.pack('<s', b'\0'))
if __name__ == '__main__':
sys.exit(Main(sys.argv))