diff --git a/resources/about_credits.tmpl b/resources/about_credits.tmpl
new file mode 100644
index 00000000..00b4d2ef
--- /dev/null
+++ b/resources/about_credits.tmpl
@@ -0,0 +1,101 @@
+
+
+
+
+Credits
+
+
+
+Credits
+Print
+
+{{entries}}
+
+
+
+
diff --git a/resources/about_credits_entry.tmpl b/resources/about_credits_entry.tmpl
new file mode 100644
index 00000000..a4fa4e1c
--- /dev/null
+++ b/resources/about_credits_entry.tmpl
@@ -0,0 +1,8 @@
+
diff --git a/script/create-dist b/script/create-dist
index 8fd286aa..19fd0213 100755
--- a/script/create-dist
+++ b/script/create-dist
@@ -310,23 +310,26 @@ def run_ar_combine(filename, target_dir):
def generate_licenses():
- licenses_py = os.path.join(SRC_DIR, 'tools', 'licenses.py')
- output = subprocess.check_output([licenses_py, 'credits'])
+ file_template = os.path.join(SOURCE_ROOT, 'resources', 'about_credits.tmpl')
+ entry_template = os.path.join(SOURCE_ROOT, 'resources',
+ 'about_credits_entry.tmpl')
+ licenses_py = os.path.join(SOURCE_ROOT, 'tools', 'licenses.py')
target = os.path.join(MAIN_DIR, 'LICENSES.chromium.html')
- with open(target, 'w') as f:
- f.write(output)
+ subprocess.check_call([licenses_py, 'credits', target,
+ '--file-template', file_template,
+ '--entry-template', entry_template])
def create_zip():
print 'Zipping shared_library builds...'
p = os.path.join(SOURCE_ROOT, 'libchromiumcontent.zip')
- make_zip(MAIN_DIR, ['LICENSES.chromium.html', 'src', 'shared_library'], p)
+ make_zip(MAIN_DIR, ['src', 'shared_library'], ['LICENSES.chromium.html'], p)
print 'Zipping static_library builds...'
p = os.path.join(SOURCE_ROOT, 'libchromiumcontent-static.zip')
- make_zip(MAIN_DIR, ['static_library'], p)
+ make_zip(MAIN_DIR, ['static_library'], [], p)
-def make_zip(src, dirs, target):
+def make_zip(src, dirs, files, target):
safe_unlink(target)
with scoped_cwd(src):
zip_file = zipfile.ZipFile(target, 'w', zipfile.ZIP_DEFLATED)
@@ -334,6 +337,8 @@ def make_zip(src, dirs, target):
for root, _, filenames in os.walk(dirname):
for f in filenames:
zip_file.write(os.path.join(root, f))
+ for f in files:
+ zip_file.write(f)
zip_file.close();
diff --git a/tools/licenses.py b/tools/licenses.py
new file mode 100755
index 00000000..605aa6ab
--- /dev/null
+++ b/tools/licenses.py
@@ -0,0 +1,511 @@
+#!/usr/bin/env python
+# Copyright (c) 2012 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.
+
+"""Utility for checking and processing licensing information in third_party
+directories.
+
+Usage: licenses.py
+
+Commands:
+ scan scan third_party directories, verifying that we have licensing info
+ credits generate about:credits on stdout
+
+(You can also import this as a module.)
+"""
+
+import argparse
+import cgi
+import os
+import sys
+
+# Paths from the root of the tree to directories to skip.
+PRUNE_PATHS = set([
+ # Same module occurs in crypto/third_party/nss and net/third_party/nss, so
+ # skip this one.
+ os.path.join('third_party','nss'),
+
+ # Placeholder directory only, not third-party code.
+ os.path.join('third_party','adobe'),
+
+ # Apache 2.0 license. See crbug.com/140478
+ os.path.join('third_party','bidichecker'),
+
+ # Build files only, not third-party code.
+ os.path.join('third_party','widevine'),
+
+ # Only binaries, used during development.
+ os.path.join('third_party','valgrind'),
+
+ # Used for development and test, not in the shipping product.
+ os.path.join('build','secondary'),
+ os.path.join('third_party','bison'),
+ os.path.join('third_party','blanketjs'),
+ os.path.join('third_party','cygwin'),
+ os.path.join('third_party','gles2_conform'),
+ os.path.join('third_party','gnu_binutils'),
+ os.path.join('third_party','gold'),
+ os.path.join('third_party','gperf'),
+ os.path.join('third_party','lighttpd'),
+ os.path.join('third_party','llvm'),
+ os.path.join('third_party','llvm-build'),
+ os.path.join('third_party','mingw-w64'),
+ os.path.join('third_party','nacl_sdk_binaries'),
+ os.path.join('third_party','pefile'),
+ os.path.join('third_party','perl'),
+ os.path.join('third_party','psyco_win32'),
+ os.path.join('third_party','pylib'),
+ os.path.join('third_party','pywebsocket'),
+ os.path.join('third_party','qunit'),
+ os.path.join('third_party','sinonjs'),
+ os.path.join('third_party','syzygy'),
+ os.path.join('tools', 'profile_chrome', 'third_party'),
+
+ # Chromium code in third_party.
+ os.path.join('third_party','fuzzymatch'),
+ os.path.join('tools', 'swarming_client'),
+
+ # Stuff pulled in from chrome-internal for official builds/tools.
+ os.path.join('third_party', 'clear_cache'),
+ os.path.join('third_party', 'gnu'),
+ os.path.join('third_party', 'googlemac'),
+ os.path.join('third_party', 'pcre'),
+ os.path.join('third_party', 'psutils'),
+ os.path.join('third_party', 'sawbuck'),
+ # See crbug.com/350472
+ os.path.join('chrome', 'browser', 'resources', 'chromeos', 'quickoffice'),
+ # Chrome for Android proprietary code.
+ os.path.join('clank'),
+
+ # Redistribution does not require attribution in documentation.
+ os.path.join('third_party','directxsdk'),
+ os.path.join('third_party','platformsdk_win2008_6_1'),
+ os.path.join('third_party','platformsdk_win7'),
+
+ # For testing only, presents on some bots.
+ os.path.join('isolate_deps_dir'),
+])
+
+# Directories we don't scan through.
+VCS_METADATA_DIRS = ('.svn', '.git')
+PRUNE_DIRS = (VCS_METADATA_DIRS +
+ ('out', 'Debug', 'Release', # build files
+ 'layout_tests')) # lots of subdirs
+
+ADDITIONAL_PATHS = (
+ os.path.join('breakpad'),
+ os.path.join('chrome', 'common', 'extensions', 'docs', 'examples'),
+ os.path.join('chrome', 'test', 'chromeos', 'autotest'),
+ os.path.join('chrome', 'test', 'data'),
+ os.path.join('native_client'),
+ os.path.join('net', 'tools', 'spdyshark'),
+ os.path.join('sdch', 'open-vcdiff'),
+ os.path.join('testing', 'gmock'),
+ os.path.join('testing', 'gtest'),
+ os.path.join('tools', 'grit'),
+ os.path.join('tools', 'gyp'),
+ os.path.join('tools', 'page_cycler', 'acid3'),
+ os.path.join('url', 'third_party', 'mozilla'),
+ os.path.join('v8'),
+ # Fake directories to include the strongtalk and fdlibm licenses.
+ os.path.join('v8', 'strongtalk'),
+ os.path.join('v8', 'fdlibm'),
+)
+
+
+# Directories where we check out directly from upstream, and therefore
+# can't provide a README.chromium. Please prefer a README.chromium
+# wherever possible.
+SPECIAL_CASES = {
+ os.path.join('native_client'): {
+ "Name": "native client",
+ "URL": "http://code.google.com/p/nativeclient",
+ "License": "BSD",
+ },
+ os.path.join('sdch', 'open-vcdiff'): {
+ "Name": "open-vcdiff",
+ "URL": "http://code.google.com/p/open-vcdiff",
+ "License": "Apache 2.0, MIT, GPL v2 and custom licenses",
+ "License Android Compatible": "yes",
+ },
+ os.path.join('testing', 'gmock'): {
+ "Name": "gmock",
+ "URL": "http://code.google.com/p/googlemock",
+ "License": "BSD",
+ "License File": "NOT_SHIPPED",
+ },
+ os.path.join('testing', 'gtest'): {
+ "Name": "gtest",
+ "URL": "http://code.google.com/p/googletest",
+ "License": "BSD",
+ "License File": "NOT_SHIPPED",
+ },
+ os.path.join('third_party', 'angle'): {
+ "Name": "Almost Native Graphics Layer Engine",
+ "URL": "http://code.google.com/p/angleproject/",
+ "License": "BSD",
+ },
+ os.path.join('third_party', 'cros_system_api'): {
+ "Name": "Chromium OS system API",
+ "URL": "http://www.chromium.org/chromium-os",
+ "License": "BSD",
+ # Absolute path here is resolved as relative to the source root.
+ "License File": "/LICENSE.chromium_os",
+ },
+ os.path.join('third_party', 'lss'): {
+ "Name": "linux-syscall-support",
+ "URL": "http://code.google.com/p/linux-syscall-support/",
+ "License": "BSD",
+ "License File": "/LICENSE",
+ },
+ os.path.join('third_party', 'ots'): {
+ "Name": "OTS (OpenType Sanitizer)",
+ "URL": "http://code.google.com/p/ots/",
+ "License": "BSD",
+ },
+ os.path.join('third_party', 'pdfium'): {
+ "Name": "PDFium",
+ "URL": "http://code.google.com/p/pdfium/",
+ "License": "BSD",
+ },
+ os.path.join('third_party', 'pdfsqueeze'): {
+ "Name": "pdfsqueeze",
+ "URL": "http://code.google.com/p/pdfsqueeze/",
+ "License": "Apache 2.0",
+ "License File": "COPYING",
+ },
+ os.path.join('third_party', 'ppapi'): {
+ "Name": "ppapi",
+ "URL": "http://code.google.com/p/ppapi/",
+ },
+ os.path.join('third_party', 'scons-2.0.1'): {
+ "Name": "scons-2.0.1",
+ "URL": "http://www.scons.org",
+ "License": "MIT",
+ "License File": "NOT_SHIPPED",
+ },
+ os.path.join('third_party', 'trace-viewer'): {
+ "Name": "trace-viewer",
+ "URL": "http://code.google.com/p/trace-viewer",
+ "License": "BSD",
+ "License File": "NOT_SHIPPED",
+ },
+ os.path.join('third_party', 'v8-i18n'): {
+ "Name": "Internationalization Library for v8",
+ "URL": "http://code.google.com/p/v8-i18n/",
+ "License": "Apache 2.0",
+ },
+ os.path.join('third_party', 'WebKit'): {
+ "Name": "WebKit",
+ "URL": "http://webkit.org/",
+ "License": "BSD and GPL v2",
+ # Absolute path here is resolved as relative to the source root.
+ "License File": "/third_party/WebKit/LICENSE_FOR_ABOUT_CREDITS",
+ },
+ os.path.join('third_party', 'webpagereplay'): {
+ "Name": "webpagereplay",
+ "URL": "http://code.google.com/p/web-page-replay",
+ "License": "Apache 2.0",
+ "License File": "NOT_SHIPPED",
+ },
+ os.path.join('tools', 'grit'): {
+ "Name": "grit",
+ "URL": "http://code.google.com/p/grit-i18n",
+ "License": "BSD",
+ "License File": "NOT_SHIPPED",
+ },
+ os.path.join('tools', 'gyp'): {
+ "Name": "gyp",
+ "URL": "http://code.google.com/p/gyp",
+ "License": "BSD",
+ "License File": "NOT_SHIPPED",
+ },
+ os.path.join('v8'): {
+ "Name": "V8 JavaScript Engine",
+ "URL": "http://code.google.com/p/v8",
+ "License": "BSD",
+ },
+ os.path.join('v8', 'strongtalk'): {
+ "Name": "Strongtalk",
+ "URL": "http://www.strongtalk.org/",
+ "License": "BSD",
+ # Absolute path here is resolved as relative to the source root.
+ "License File": "/v8/LICENSE.strongtalk",
+ },
+ os.path.join('v8', 'fdlibm'): {
+ "Name": "fdlibm",
+ "URL": "http://www.netlib.org/fdlibm/",
+ "License": "Freely Distributable",
+ # Absolute path here is resolved as relative to the source root.
+ "License File" : "/v8/src/third_party/fdlibm/LICENSE",
+ "License Android Compatible" : "yes",
+ },
+ os.path.join('third_party', 'khronos_glcts'): {
+ # These sources are not shipped, are not public, and it isn't
+ # clear why they're tripping the license check.
+ "Name": "khronos_glcts",
+ "URL": "http://no-public-url",
+ "License": "Khronos",
+ "License File": "NOT_SHIPPED",
+ },
+ os.path.join('tools', 'telemetry', 'third_party', 'gsutil'): {
+ "Name": "gsutil",
+ "URL": "https://cloud.google.com/storage/docs/gsutil",
+ "License": "Apache 2.0",
+ "License File": "NOT_SHIPPED",
+ },
+}
+
+# Special value for 'License File' field used to indicate that the license file
+# should not be used in about:credits.
+NOT_SHIPPED = "NOT_SHIPPED"
+
+
+class LicenseError(Exception):
+ """We raise this exception when a directory's licensing info isn't
+ fully filled out."""
+ pass
+
+def AbsolutePath(path, filename, root):
+ """Convert a path in README.chromium to be absolute based on the source
+ root."""
+ if filename.startswith('/'):
+ # Absolute-looking paths are relative to the source root
+ # (which is the directory we're run from).
+ absolute_path = os.path.join(root, filename[1:])
+ else:
+ absolute_path = os.path.join(root, path, filename)
+ if os.path.exists(absolute_path):
+ return absolute_path
+ return None
+
+def ParseDir(path, root, require_license_file=True, optional_keys=None):
+ """Examine a third_party/foo component and extract its metadata."""
+
+ # Parse metadata fields out of README.chromium.
+ # We examine "LICENSE" for the license file by default.
+ metadata = {
+ "License File": "LICENSE", # Relative path to license text.
+ "Name": None, # Short name (for header on about:credits).
+ "URL": None, # Project home page.
+ "License": None, # Software license.
+ }
+
+ if optional_keys is None:
+ optional_keys = []
+
+ if path in SPECIAL_CASES:
+ metadata.update(SPECIAL_CASES[path])
+ else:
+ # Try to find README.chromium.
+ readme_path = os.path.join(root, path, 'README.chromium')
+ if not os.path.exists(readme_path):
+ raise LicenseError("missing README.chromium or licenses.py "
+ "SPECIAL_CASES entry")
+
+ for line in open(readme_path):
+ line = line.strip()
+ if not line:
+ break
+ for key in metadata.keys() + optional_keys:
+ field = key + ": "
+ if line.startswith(field):
+ metadata[key] = line[len(field):]
+
+ # Check that all expected metadata is present.
+ for key, value in metadata.iteritems():
+ if not value:
+ raise LicenseError("couldn't find '" + key + "' line "
+ "in README.chromium or licences.py "
+ "SPECIAL_CASES")
+
+ # Special-case modules that aren't in the shipping product, so don't need
+ # their license in about:credits.
+ if metadata["License File"] != NOT_SHIPPED:
+ # Check that the license file exists.
+ for filename in (metadata["License File"], "COPYING"):
+ license_path = AbsolutePath(path, filename, root)
+ if license_path is not None:
+ break
+
+ if require_license_file and not license_path:
+ raise LicenseError("License file not found. "
+ "Either add a file named LICENSE, "
+ "import upstream's COPYING if available, "
+ "or add a 'License File:' line to "
+ "README.chromium with the appropriate path.")
+ metadata["License File"] = license_path
+
+ return metadata
+
+
+def ContainsFiles(path, root):
+ """Determines whether any files exist in a directory or in any of its
+ subdirectories."""
+ for _, dirs, files in os.walk(os.path.join(root, path)):
+ if files:
+ return True
+ for vcs_metadata in VCS_METADATA_DIRS:
+ if vcs_metadata in dirs:
+ dirs.remove(vcs_metadata)
+ return False
+
+
+def FilterDirsWithFiles(dirs_list, root):
+ # If a directory contains no files, assume it's a DEPS directory for a
+ # project not used by our current configuration and skip it.
+ return [x for x in dirs_list if ContainsFiles(x, root)]
+
+
+def FindThirdPartyDirs(prune_paths, root):
+ """Find all third_party directories underneath the source root."""
+ third_party_dirs = set()
+ for path, dirs, files in os.walk(root):
+ path = path[len(root)+1:] # Pretty up the path.
+
+ if path in prune_paths:
+ dirs[:] = []
+ continue
+
+ # Prune out directories we want to skip.
+ # (Note that we loop over PRUNE_DIRS so we're not iterating over a
+ # list that we're simultaneously mutating.)
+ for skip in PRUNE_DIRS:
+ if skip in dirs:
+ dirs.remove(skip)
+
+ if os.path.basename(path) == 'third_party':
+ # Add all subdirectories that are not marked for skipping.
+ for dir in dirs:
+ dirpath = os.path.join(path, dir)
+ if dirpath not in prune_paths:
+ third_party_dirs.add(dirpath)
+
+ # Don't recurse into any subdirs from here.
+ dirs[:] = []
+ continue
+
+ # Don't recurse into paths in ADDITIONAL_PATHS, like we do with regular
+ # third_party/foo paths.
+ if path in ADDITIONAL_PATHS:
+ dirs[:] = []
+
+ for dir in ADDITIONAL_PATHS:
+ if dir not in prune_paths:
+ third_party_dirs.add(dir)
+
+ return third_party_dirs
+
+
+def FindThirdPartyDirsWithFiles(root):
+ third_party_dirs = FindThirdPartyDirs(PRUNE_PATHS, root)
+ return FilterDirsWithFiles(third_party_dirs, root)
+
+
+def ScanThirdPartyDirs(root=None):
+ """Scan a list of directories and report on any problems we find."""
+ if root is None:
+ root = os.getcwd()
+ third_party_dirs = FindThirdPartyDirsWithFiles(root)
+
+ errors = []
+ for path in sorted(third_party_dirs):
+ try:
+ metadata = ParseDir(path, root)
+ except LicenseError, e:
+ errors.append((path, e.args[0]))
+ continue
+
+ for path, error in sorted(errors):
+ print path + ": " + error
+
+ return len(errors) == 0
+
+
+def GenerateCredits(file_template_file, entry_template_file, output_file):
+ """Generate about:credits."""
+
+ def EvaluateTemplate(template, env, escape=True):
+ """Expand a template with variables like {{foo}} using a
+ dictionary of expansions."""
+ for key, val in env.items():
+ if escape:
+ val = cgi.escape(val)
+ template = template.replace('{{%s}}' % key, val)
+ return template
+
+ root = os.path.join(os.path.dirname(__file__), '..', 'vendor', 'chromium',
+ 'src')
+ third_party_dirs = FindThirdPartyDirs(PRUNE_PATHS, root)
+
+ if not file_template_file:
+ file_template_file = os.path.join(root, 'chrome', 'browser',
+ 'resources', 'about_credits.tmpl')
+ if not entry_template_file:
+ entry_template_file = os.path.join(root, 'chrome', 'browser',
+ 'resources',
+ 'about_credits_entry.tmpl')
+
+ entry_template = open(entry_template_file).read()
+ entries = []
+ for path in third_party_dirs:
+ try:
+ metadata = ParseDir(path, root)
+ except LicenseError:
+ # TODO(phajdan.jr): Convert to fatal error (http://crbug.com/39240).
+ continue
+ if metadata['License File'] == NOT_SHIPPED:
+ continue
+ env = {
+ 'name': metadata['Name'],
+ 'url': metadata['URL'],
+ 'license': open(metadata['License File'], 'rb').read(),
+ }
+ entry = {
+ 'name': metadata['Name'],
+ 'content': EvaluateTemplate(entry_template, env),
+ }
+ entries.append(entry)
+
+ entries.sort(key=lambda entry: (entry['name'], entry['content']))
+ entries_contents = '\n'.join([entry['content'] for entry in entries])
+ file_template = open(file_template_file).read()
+ template_contents = ""
+ template_contents += EvaluateTemplate(file_template,
+ {'entries': entries_contents},
+ escape=False)
+
+ if output_file:
+ with open(output_file, 'w') as output:
+ output.write(template_contents)
+ else:
+ print template_contents
+
+ return True
+
+
+def main():
+ parser = argparse.ArgumentParser()
+ parser.add_argument('--file-template',
+ help='Template HTML to use for the license page.')
+ parser.add_argument('--entry-template',
+ help='Template HTML to use for each license.')
+ parser.add_argument('command', choices=['help', 'scan', 'credits'])
+ parser.add_argument('output_file', nargs='?')
+ args = parser.parse_args()
+
+ if args.command == 'scan':
+ if not ScanThirdPartyDirs():
+ return 1
+ elif args.command == 'credits':
+ if not GenerateCredits(args.file_template, args.entry_template,
+ args.output_file):
+ return 1
+ else:
+ print __doc__
+ return 1
+
+
+if __name__ == '__main__':
+ sys.exit(main())