#!/usr/bin/python
#
# Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
#

"""Script that attempts to generate an HTML file containing license
information and homepage links for all installed packages.

WARNING: this script in its current form is not finished or considered
production quality/code style compliant. This is an intermediate checkin
to allow for incremental cleanups and improvements that will make it
production quality.

Usage:
For this script to work, you must have built the architecture
this is being run against, _after_ you've last run repo sync.
Otherwise, it will query newer source code and then fail to work on packages
that are out of date in your build.

Recommended build:
  cros_sdk
  export board=x86-alex
  sudo rm -rf /build/$board
  cd ~/trunk/src/scripts
  # TODO: setup_board should generate static list of packages that can be used
  # by this script.
  ./setup_board --board=$board
  # If you wonder why we need to build
  # chromeos just to run emerge -p -v chromeos-base/chromeos on it, we don't.
  # However, later we run ebuild unpack, and this will apply patches and run
  # configure. Configure will fail due to aclocal macros missing in
  # /build/x86-alex/usr/share/aclocal (those are generated during build).
  # This will take about 10mn on a Z620.
  ./build_packages --board=$board --nowithautotest --nowithtest --nowithdev \
                   --nowithfactory
  cd ~/trunk/chromite/licensing
  %(prog)s [--debug] --board $board -o out.html 2>&1 | tee output.sav

You can check the licenses and/or generate a HTML file for a list of
packages using --package or -p:
  %(prog)s --package "dev-libs/libatomic_ops-7.2d" --package
  "net-misc/wget-1.14" --board $board -o out.html

By default, when no package is specified, this script processes all
packages for the board. The output HTML file is meant to update
http://src.chromium.org/viewvc/chrome/trunk/src/chrome/browser/resources/ +
  chromeos/about_os_credits.html?view=log
(gclient config svn://svn.chromium.org/chrome/trunk/src)
For an example CL, see https://codereview.chromium.org/13496002/

The detailed process is listed below.

* Check out the branch you intend to generate the HTML file for. Use
  the internal manifest for this purpose.
    repo init -b <branch_name> -u <URL>

  The list of branches (e.g. release-R33-5116.B) are available here:
  https://chromium.googlesource.com/chromiumos/manifest/+refs

* Generate the HTML file by following the steps mentioned
  previously. Check whether your changes are valid with:
    bin/diff_license_html output.html-M33 output.html-M34
  and review the diff.

* Update the about_os_credits.html in the svn repository. Create a CL
  and upload it for review.
    gcl change <change_name>
    gcl upload <change_name>

  When uploading, you may get a warning for file being too large to
  upload. In this case, your CL can still be reviewed. Always include
  the diff in your commit message so that the reviwers know what the
  changes are. You can add reviewers on the reivew page by clicking on
  "Edit issue".  (A quick reference:
  http://www.chromium.org/developers/quick-reference)

* After receiving LGTMs, commit your change with 'gcl commit <change_name>'.

If you don't get this in before the freeze window, it'll need to be merged into
the branch being released, which is done by adding a Merge-Requested label to
Iteration-xx in the tracking bug.
Once it's been updated to "Merge-Approved" by a TPM, please merge into the
required release branch. You can ask karen@ for merge approve help.
Example: http://crbug.com/221281

Usage: %(prog)s [opts] <board>

"""

import cgi
import codecs
import logging
import os
import re
import tempfile

from chromite.buildbot import constants
from chromite.buildbot import portage_utilities
from chromite.lib import commandline
from chromite.lib import cros_build_lib
from chromite.lib import osutils


debug = False

STOCK_LICENSE_DIRS = [
    os.path.join(constants.SOURCE_ROOT, 'src/third_party/portage/licenses'),
    os.path.join(constants.SOURCE_ROOT,
                 'src/third_party/portage-stable/licenses'),
]

# There are licenses for custom software we got and isn't part of
# upstream gentoo.
CUSTOM_LICENSE_DIRS = [
    os.path.join(constants.SOURCE_ROOT,
                 'src/third_party/chromiumos-overlay/licenses'),
]

COPYRIGHT_ATTRIBUTION_DIR = (
    os.path.join(
        constants.SOURCE_ROOT,
        'src/third_party/chromiumos-overlay/licenses/copyright-attribution'))

# Virtual packages don't need to have a license and often don't, so we skip them
# chromeos-base contains google platform packages that are covered by the
# general license at top of tree, so we skip those too.
SKIPPED_CATEGORIES = [
    'chromeos-base',  # TODO: it wouldn't hurt to remove this and make sure our
                      # packages all come with a BSD-Google license.
    'virtual',
]

SKIPPED_PACKAGES = [
    # Fix these packages by adding a real license in the code.
    # You should not skip packages just because the license scraping doesn't
    # work. Stick those special cases into PACKAGE_LICENSES.
    # Packages should only be here because they are sub/split packages already
    # covered by the license of the main package.

    # These are Chrome-OS-specific packages, copyright BSD-Google
    'sys-kernel/chromeos-kernel',  # already manually credit Linux

    # These have been split across several packages, so we skip listing the
    # individual components (and just list the main package instead).
    'app-editors/vim-core',
    'x11-apps/mesa-progs',

    # Portage metapackage.
    'x11-base/xorg-drivers',

    # These are covered by app-i18n/ibus-mozc (BSD, copyright Google).
    'app-i18n/ibus-mozc-chewing',
    'app-i18n/ibus-mozc-hangul',
    'app-i18n/ibus-mozc-pinyin',

    # Those have License: Proprietary in the ebuild.
    # FIXME? Should the code detect 'Proprietary' and exclude the packages, or
    # list them with a license that says 'Proprietary'?
    'app-i18n/GoogleChineseInput-cangjie',
    'app-i18n/GoogleChineseInput-pinyin',
    'app-i18n/GoogleChineseInput-wubi',
    'app-i18n/GoogleChineseInput-zhuyin',
    'app-i18n/GoogleKoreanInput',

    # These are all X.org sub-packages; shouldn't be any need to list them
    # individually.
    'media-fonts/encodings',
    'x11-apps/iceauth',
    'x11-apps/intel-gpu-tools',
    'x11-apps/mkfontdir',
    'x11-apps/rgb',
    'x11-apps/setxkbmap',
    'x11-apps/xauth',
    'x11-apps/xcursorgen',
    'x11-apps/xdpyinfo',
    'x11-apps/xdriinfo',
    'x11-apps/xev',
    'x11-apps/xgamma',
    'x11-apps/xhost',
    'x11-apps/xinit',
    'x11-apps/xinput',
    'x11-apps/xkbcomp',
    'x11-apps/xlsatoms',
    'x11-apps/xlsclients',
    'x11-apps/xmodmap',
    'x11-apps/xprop',
    'x11-apps/xrandr',
    'x11-apps/xrdb',
    'x11-apps/xset',
    'x11-apps/xset-mini',
    'x11-apps/xwininfo',
    'x11-base/xorg-server',
    'x11-drivers/xf86-input-evdev',
    'x11-drivers/xf86-input-keyboard',
    'x11-drivers/xf86-input-mouse',
    'x11-drivers/xf86-input-synaptics',
    'x11-drivers/xf86-video-intel',
    'x11-drivers/xf86-video-vesa',
    'x11-drivers/xf86-video-vmware',
    'x11-libs/libICE',
    'x11-libs/libSM',
    'x11-libs/libX11',
    'x11-libs/libXScrnSaver',
    'x11-libs/libXau',
    'x11-libs/libXcomposite',
    'x11-libs/libXcursor',
    'x11-libs/libXdamage',
    'x11-libs/libXdmcp',
    'x11-libs/libXext',
    'x11-libs/libXfixes',
    'x11-libs/libXfont',
    'x11-libs/libXfontcache',
    'x11-libs/libXft',
    'x11-libs/libXi',
    'x11-libs/libXinerama',
    'x11-libs/libXmu',
    'x11-libs/libXp',
    'x11-libs/libXrandr',
    'x11-libs/libXrender',
    'x11-libs/libXres',
    'x11-libs/libXt',
    'x11-libs/libXtst',
    'x11-libs/libXv',
    'x11-libs/libXvMC',
    'x11-libs/libXxf86vm',
    'x11-libs/libdrm',
    'x11-libs/libfontenc',
    'x11-libs/libpciaccess',
    'x11-libs/libxkbfile',
    'x11-libs/libxkbui',
    'x11-libs/pixman',
    'x11-libs/xtrans',
    'x11-misc/util-macros',
    'x11-misc/xbitmaps',
    'x11-proto/bigreqsproto',
    'x11-proto/compositeproto',
    'x11-proto/damageproto',
    'x11-proto/dri2proto',
    'x11-proto/fixesproto',
    'x11-proto/fontcacheproto',
    'x11-proto/fontsproto',
    'x11-proto/inputproto',
    'x11-proto/kbproto',
    'x11-proto/printproto',
    'x11-proto/randrproto',
    'x11-proto/recordproto',
    'x11-proto/renderproto',
    'x11-proto/resourceproto',
    'x11-proto/scrnsaverproto',
    'x11-proto/trapproto',
    'x11-proto/videoproto',
    'x11-proto/xcmiscproto',
    'x11-proto/xextproto',
    'x11-proto/xf86bigfontproto',
    'x11-proto/xf86dgaproto',
    'x11-proto/xf86driproto',
    'x11-proto/xf86rushproto',
    'x11-proto/xf86vidmodeproto',
    'x11-proto/xineramaproto',
    'x11-proto/xproto',
]

# TODO(merlin): replace matching with regex matching to simplify this
# Matching is done in lowercase, you MUST give lowercase names.
LICENSE_FILENAMES = [
    'copying',
    'copyright',
    'ipa_font_license_agreement_v1.0.txt',  # used by ja-ipafonts
    'licence',        # used by openssh
    'license',
    'license.txt',    # used by hdparm, NumPy, glew
    'licensing.txt',  # used by libatomic_ops
]
# FIXME, allow for these licenses to be regexes instead. For instance,
# sys-apps/util-linux-2.21.2-r1 should show Documentation/licenses/COPYING.BSD-3

# These are _temporary_ license mappings for packages that do not have a valid
# shared/custom license, or LICENSE file we can use.
# Once this script runs earlier (during the package build process), it will
# block new source without a LICENSE file if the ebuild contains a license
# that requires copyright assignment (BSD and friends).
# At that point, new packages will get fixed to include LICENSE instead of
# adding workaround mappings like those below.
# The way you now fix copyright attribution cases create a custom file with the
# right license directly in COPYRIGHT_ATTRIBUTION_DIR.
PACKAGE_LICENSES = {
    # Example of what we used to have here. The code that uses this dictionary
    # will be removed in the near future, so don't rely on adding anything here
    # longterm.
    # 'sys-libs/ncurses': ['ncurses'],
}

# Any license listed list here found in the ebuild will make the code look for
# license files inside the package source code in order to get copyright
# attribution from them.
COPYRIGHT_ATTRIBUTION_LICENSES = [
    'BSD',    # requires distribution of copyright notice
    'BSD-2',  # so does BSD-2 http://opensource.org/licenses/BSD-2-Clause
    'BSD-3',  # and BSD-3? http://opensource.org/licenses/BSD-3-Clause
    'BSD-4',  # and 4?
    'BSD-with-attribution',
    'MIT',
    'MIT-with-advertising',
    'Old-MIT',
]

# The following licenses are not invalid or to show as a less helpful stock
# license, but it's better to look in the source code for a more specific
# license if there is one, but not an error if no better one is found.
# Note that you don't want to set just anything here since any license here
# will be included once in stock form and a second time in custom form if
# found (there is no good way to know that a license we found on disk is the
# better version of the stock version, so we show both).
LOOK_IN_SOURCE_LICENSES = [
    'as-is',  # The stock license is very vague, source always has more details.
    'PSF-2',  # The custom license in python is more complete than the template.

    # As far as I know, we have no requirement to do copyright attribution for
    # these licenses, but the license included in the code has slightly better
    # information than the stock Gentoo one (including copyright attribution).
    'BZIP2',     # Single use license, do copyright attribution.
    'OFL',       # Almost single use license, do copyright attribution.
    'OFL-1.1',   # Almost single use license, do copyright attribution.
    'UoI-NCSA',  # Only used by NSCA, might as well show their custom copyright.
]

# This used to provide overrides. I can't find a valid reason to add any more
# here, though.
PACKAGE_HOMEPAGES = {
    # Example:
    # 'x11-proto/glproto': ['http://www.x.org/'],
}

# These are tokens found in LICENSE= in an ebuild that aren't licenses we
# can actually read from disk.
# You should not use this to blacklist real licenses.
LICENCES_IGNORE = [
    ')',              # Ignore OR tokens from LICENSE="|| ( LGPL-2.1 MPL-1.1 )"
    '(',
    '||',
]

TMPL = 'about_credits.tmpl'
ENTRY_TMPL = 'about_credits_entry.tmpl'
SHARED_LICENSE_TMPL = 'about_credits_shared_license_entry.tmpl'


def GetLicenseTypesFromEbuild(ebuild_path):
  """Returns a list of license types from the ebuild file.

  This function does not always return the correct list, but it is
  faster than using portageq for not having to access chroot. It is
  intended to be used for tasks such as presubmission checks.
  """
  ebuild_env_tmpl = """
has() { [[ " ${*:2} " == *" $1 "* ]]; }
inherit() {
  local overlay_list="%(overlay_list)s"
  local eclass overlay f
  for eclass; do
    has ${eclass} ${_INHERITED_} && continue
    _INHERITED_+=" ${eclass}"
    for overlay in %(overlay_list)s; do
      f="${overlay}/eclass/${eclass}.eclass"
      if [[ -e ${f} ]]; then
        source "${f}"
        break
      fi
     done
  done
}
source %(ebuild)s"""

  # TODO: the overlay_list hard-coded here should be changed to look
  # at the current overlay, and then the master overlays. E.g. for an
  # ebuild file in overlay-parrot, we will look at parrot overlay
  # first, and then look at portage-stable and chromiumos, which are
  # listed as masters in overlay-parrot/metadata/layout.conf.
  tmpl_env = {
      'ebuild': ebuild_path,
      'overlay_list': '%s %s' % (
          os.path.join(constants.SOURCE_ROOT,
                       'src/third_party/chromiumos-overlay'),
          os.path.join(constants.SOURCE_ROOT,
                       'src/third_party/portage-stable'))
  }

  with tempfile.NamedTemporaryFile(bufsize=0) as f:
    osutils.WriteFile(f.name, ebuild_env_tmpl % tmpl_env)
    env = osutils.SourceEnvironment(
        f.name, whitelist=['LICENSE'], ifs=' ', multiline=True)

  if not env.get('LICENSE'):
    raise ValueError('No LICENSE found in the ebuild.')
  if re.search(r'[,;]', env['LICENSE']):
    raise ValueError(
        'LICENSE field in the ebuild should be whitespace-limited.')

  return env['LICENSE'].split()


class PackageLicenseError(Exception):
  """Thrown if something fails while getting license information for a package.

  This will cause the processing to error in the end.
  """


class PackageSkipped(Exception):
  """Non error to exclude packages from license processing."""


class PackageInfo(object):
  """Package info containers, mostly for storing licenses."""

  def __init__(self, board):

    self.board = board
    self.revision = None

    # Array of scanned license texts.
    self.license_text_scanned = []

    self.category = None
    self.name = None
    self.version = None

    # Looks something like this
    # /mnt/host/source/src/
    #           third_party/portage-stable/net-misc/rsync/rsync-3.0.8.ebuild
    self.ebuild_path = None
    # dirname of ebuild_path.
    self.ebuild_dir = None

    # Array of license names retrieved from ebuild or override in this code.
    self.ebuild_license_names = []
    self.description = None
    self.homepages = []
    # This contains licenses names we can read from Gentoo or custom licenses.
    # These are supposed to be shared licenses (i.e. licenses referenced by
    # more then one package), but after all processing, we may find out that
    # some are only used once and they get taken out of the shared pool and
    # pasted directly in the sole package that was using them (see
    # GenerateHTMLLicenseOutput).
    self.license_names = set()

    # We set this if the ebuild has a BSD/MIT like license that requires
    # scanning for a LICENSE file in the source code, or a static mapping
    # in PACKAGE_LICENSES. Not finding one once this is set, is fatal.
    self.need_copyright_attribution = False
    # This flag just says we'd like to include licenses from the source, but
    # not finding any is not fatal.
    self.scan_source_for_licenses = False

  @property
  def fullnamerev(self):
    s = '%s-%s' % (self.fullname, self.version)
    if self.revision:
      s += '-r%s' % self.revision
    return s

  @property
  def fullname(self):
    return '%s/%s' % (self.category, self.name)

  def _RunEbuildPhases(self, phases):
    """Run a list of ebuild phases on an ebuild.

    Args:
      phases: list of phases like ['clean', 'fetch'] or ['unpack'].

    Returns:
      ebuild command output
    """

    return cros_build_lib.RunCommand(
        ['ebuild-%s' % self.board, self.ebuild_path] + phases, print_cmd=debug,
        redirect_stdout=True)

  def _GetOverrideLicense(self):
    """Look in COPYRIGHT_ATTRIBUTION_DIR for license with copyright attribution.

    For dev-util/bsdiff-4.3-r5, the code will look for
    dev-util/bsdiff-4.3-r5
    dev-util/bsdiff-4.3
    dev-util/bsdiff

    It is ok to have more than one bsdiff license file, and an empty file acts
    as a rubout (i.e. an empty dev-util/bsdiff-4.4 will shadow dev-util/bsdiff
    and tell the licensing code to look in the package source for a license
    instead of using dev-util/bsdiff as an override).

    Returns:
      False (no license found) or a multiline license string.
    """

    license_read = None
    # dev-util/bsdiff-4.3-r5 -> bsdiff-4.3-r5
    filename = os.path.basename(self.fullnamerev)
    license_path = os.path.join(COPYRIGHT_ATTRIBUTION_DIR,
                                os.path.dirname(self.fullnamerev))
    pv = portage_utilities.SplitPV(filename)
    pv_no_rev = '%s-%s' % (pv.package, pv.version_no_rev)
    for filename in (pv.pv, pv_no_rev, pv.package):
      file_path = os.path.join(license_path, filename)
      logging.debug("Looking for override copyright attribution license in %s",
                    file_path)
      if os.path.exists(file_path):
        # Turn
        # /../merlin/trunk/src/third_party/chromiumos-overlay/../dev-util/bsdiff
        # into
        # chromiumos-overlay/../dev-util/bsdiff
        short_dir_path = os.path.join(*file_path.rsplit(os.path.sep, 5)[1:])
        license_read = "Copyright Attribution License %s:\n\n" % short_dir_path
        license_read += ReadUnknownEncodedFile(
            file_path, "read copyright attribution license")
        break

    return license_read

  def _ExtractLicenses(self):
    """Scrounge for text licenses in the source of package we'll unpack.

    This is only called if we couldn't get usable licenses from the ebuild,
    or one of them is BSD/MIT like which forces us to look for a file with
    copyright attribution in the source code itself.

    First, we have a shortcut where we scan COPYRIGHT_ATTRIBUTION_DIR to see if
    we find a license for this package. If so, we use that.
    Typically it'll be used if the unpacked source does not have the license
    that we're required to display for copyright attribution (in some cases it's
    plain absent, in other cases, it could be in a filename we don't look for).

    Otherwise, we scan the unpacked source code for what looks like license
    files as defined in LICENSE_FILENAMES.

    Raises:
      AssertionError: on runtime errors
      PackageLicenseError: couldn't find copyright attribution file.
    """

    license_override = self._GetOverrideLicense()
    if license_override:
      self.license_text_scanned = [license_override]
      return

    self._RunEbuildPhases(['clean', 'fetch'])
    output = self._RunEbuildPhases(['unpack']).output.splitlines()
    # Output is spammy, it looks like this:
    #  * gc-7.2d.tar.gz RMD160 SHA1 SHA256 size ;-) ...                  [ ok ]
    #  * checking gc-7.2d.tar.gz ;-) ...                                 [ ok ]
    #  * Running stacked hooks for pre_pkg_setup
    #  *    sysroot_build_bin_dir ...
    #  [ ok ]
    #  * Running stacked hooks for pre_src_unpack
    #  *    python_multilib_setup ...
    #  [ ok ]
    # >>> Unpacking source...
    # >>> Unpacking gc-7.2d.tar.gz to /build/x86-alex/tmp/po/[...]tops-7.2d/work
    # >>> Source unpacked in /build/x86-alex/tmp/portage/[...]ops-7.2d/work
    # So we only keep the last 2 lines, the others we don't care about.
    output = [line for line in output if line[0:3] == ">>>" and
              line != ">>> Unpacking source..."]
    for line in output:
      logging.info(line)

    args = ['portageq-%s' % self.board, 'envvar', 'PORTAGE_TMPDIR']
    result = cros_build_lib.RunCommand(args, print_cmd=debug,
                                       redirect_stdout=True)
    tmpdir = result.output.splitlines()[0]
    # tmpdir gets something like /build/daisy/tmp/
    workdir = os.path.join(tmpdir, 'portage', self.fullnamerev, 'work')

    if not os.path.exists(workdir):
      raise AssertionError("Unpack of %s didn't create %s. Version mismatch?" %
                           (self.fullnamerev, workdir))

    # You may wonder how deep should we go?
    # In case of packages with sub-packages, it could be deep.
    # Let's just be safe and get everything we can find.
    # In the case of libatomic_ops, it's actually required to look deep
    # to find the MIT license:
    # dev-libs/libatomic_ops-7.2d/work/gc-7.2/libatomic_ops/doc/LICENSING.txt
    args = ['find', workdir, '-type', 'f']
    result = cros_build_lib.RunCommand(args, print_cmd=debug,
                                       redirect_stdout=True).output.splitlines()
    # Truncate results to look like this: swig-2.0.4/COPYRIGHT
    files = [x[len(workdir):].lstrip('/') for x in result]
    license_files = []
    for name in files:
      if os.path.basename(name).lower() in LICENSE_FILENAMES:
        license_files.append(name)

    if not license_files:
      if self.need_copyright_attribution:
        logging.error("""
%s: unable to find usable license.
Typically this will happen because the ebuild says it's MIT or BSD, but there
was no license file that this script could find to include along with a
copyright attribution (required for BSD/MIT).

If this is Google source, please change
LICENSE="BSD"
to
LICENSE="BSD-Google"

If not, go investigate the unpacked source in %s,
and find which license to assign.  Once you found it, you should copy that
license to a file under %s
(or you can modify LICENSE_FILENAMES to pickup a license file that isn't
being scraped currently).""",
                      self.fullnamerev, workdir, COPYRIGHT_ATTRIBUTION_DIR)
        raise PackageLicenseError()
      else:
        # We can get called for a license like as-is where it's preferable
        # to find a better one in the source, but not fatal if we didn't.
        logging.info("Was not able to find a better license for %s "
                     "in %s to replace the more generic one from ebuild",
                     self.fullnamerev, workdir)

    # Examples of multiple license matches:
    # dev-lang/swig-2.0.4-r1: swig-2.0.4/COPYRIGHT swig-2.0.4/LICENSE
    # dev-libs/glib-2.32.4-r1: glib-2.32.4/COPYING pkg-config-0.26/COPYING
    # dev-libs/libnl-3.2.14: libnl-doc-3.2.14/COPYING libnl-3.2.14/COPYING
    # dev-libs/libpcre-8.30-r2: pcre-8.30/LICENCE pcre-8.30/COPYING
    # dev-libs/libusb-0.1.12-r6: libusb-0.1.12/COPYING libusb-0.1.12/LICENSE
    # dev-libs/pyzy-0.1.0-r1: db/COPYING pyzy-0.1.0/COPYING
    # net-misc/strongswan-5.0.2-r4: strongswan-5.0.2/COPYING
    #                               strongswan-5.0.2/LICENSE
    # sys-process/procps-3.2.8_p11: debian/copyright procps-3.2.8/COPYING
    logging.info('License(s) for %s: %s', self.fullnamerev,
                 ' '.join(license_files))
    for license_file in sorted(license_files):
      # Joy and pink ponies. Some license_files are encoded as latin1 while
      # others are utf-8 and of course you can't know but only guess.
      license_path = os.path.join(workdir, license_file)
      license_txt = ReadUnknownEncodedFile(license_path, "Adding License")

      self.license_text_scanned += [
          "Scanned Source License %s:\n\n%s" % (license_file, license_txt)]

    # We used to clean up here, but there have been many instances where
    # looking at unpacked source to see where the licenses were, was useful
    # so let's disable this for now
    # self._RunEbuildPhases(['clean'])

  def _SetPackageInfo(self, fullnamewithrev):
    """Populate PackageInfo with package license, homepage and description.

    Some packages have static license mappings applied to them.

    self.ebuild_license_names will not be filled if the package is skipped
    or if there was an issue getting data from the ebuild.
    self.license_names will only get the licenses that we can paste
    as shared licenses.
    scan_source_for_licenses will be set if we should unpack the source to look
    for licenses
    if need_copyright_attribution is also set, not finding a license in the
    source is fatal (PackageLicenseError will get raised).

    Args:
      fullnamewithrev: e.g. dev-libs/libatomic_ops-7.2d

    Raises:
      AssertionError: on runtime errors
      PackageSkipped: if in skip list
    """

    try:
      cpv = portage_utilities.SplitCPV(fullnamewithrev)
      (self.category, self.name, self.version, self.revision) = (
          cpv.category, cpv.package, cpv.version_no_rev, cpv.rev)
    except TypeError:
      raise AssertionError("portage couldn't find %s, missing version number?" %
                           fullnamewithrev)

    if self.revision is not None:
      self.revision = str(self.revision).lstrip('r')
      if self.revision == '0':
        self.revision = None

    if self.category in SKIPPED_CATEGORIES:
      raise PackageSkipped("%s in SKIPPED_CATEGORIES, skip package" %
                           self.fullname)

    if self.fullname in SKIPPED_PACKAGES:
      raise PackageSkipped("%s in SKIPPED_PACKAGES, skip package" %
                           self.fullname)


    # By default, equery returns the latest version of the package. A
    # build may have used an older version than what is currently
    # available in the source tree (a build dependency can be pinned
    # to an older version of a package for compatibility
    # reasons). Therefore we need to tell equery that we want the
    # exact version number used in the image build as opposed to the
    # latest available in the source tree.
    args = ['equery-%s' % self.board, 'which', self.fullnamerev]
    path = cros_build_lib.RunCommand(args, print_cmd=debug,
                                     redirect_stdout=True).output.strip()
    logging.debug("%s -> %s", " ".join(args), path)
    if not path:
      raise AssertionError('GetEbuildPath for %s failed.\n'
                           'Is your tree clean? Delete %s and rebuild' %
                           (self.name,
                            cros_build_lib.GetSysroot(board=self.board)))

    if not os.access(path, os.F_OK):
      raise AssertionError("Can't access %s", path)

    self.ebuild_dir = os.path.dirname(path)
    self.ebuild_path = path

    args = ['portageq-%s' % self.board, 'metadata',
            cros_build_lib.GetSysroot(board=self.board), 'ebuild',
            self.fullnamerev, 'HOMEPAGE', 'LICENSE', 'DESCRIPTION']
    lines = cros_build_lib.RunCommand(args, print_cmd=debug,
                                      redirect_stdout=True).output.splitlines()
    # Runs:
    # portageq metadata /build/x86-alex ebuild net-misc/wget-1.12-r2 \
    #                                               HOMEPAGE LICENSE DESCRIPTION
    # Returns:
    # http://www.gnu.org/software/wget/
    # GPL-3
    # Network utility to retrieve files from the WWW

    (self.homepages,   self.ebuild_license_names, self.description) = (
     lines[0].split(), lines[1].split(),          lines[2:])

    if self.fullname in PACKAGE_HOMEPAGES:
      self.homepages = PACKAGE_HOMEPAGES[self.fullname]

    # Packages with missing licenses or licenses that need mapping (like
    # BSD/MIT) are hardcoded here:
    if self.fullname in PACKAGE_LICENSES:
      self.ebuild_license_names = PACKAGE_LICENSES[self.fullname]
      logging.info("Static license mapping for %s: %s", self.fullnamerev,
                   ",".join(self.ebuild_license_names))
    else:
      logging.info("Read licenses from ebuild for %s: %s", self.fullnamerev,
                   ",".join(self.ebuild_license_names))

  def GetLicenses(self, fullnamewithrev):
    """Get licenses from the ebuild field and the unpacked source code.

    After populating the package info and licenses, this figures
    out whether the package source should be scanned to add licenses found
    there.

    Args:
      fullnamewithrev: e.g. dev-libs/libatomic_ops-7.2d

    Raises:
      AssertionError: on runtime errors
      PackageLicenseError: couldn't find license in ebuild.
    """

    # First populate the package basic information
    self._SetPackageInfo(fullnamewithrev)

    # The ebuild license field can look like:
    # LICENSE="GPL-3 LGPL-3 Apache-2.0" (this means AND, as in all 3)
    # for third_party/portage-stable/app-admin/rsyslog/rsyslog-5.8.11.ebuild
    # LICENSE="|| ( LGPL-2.1 MPL-1.1 )"
    # for third_party/portage-stable/x11-libs/cairo/cairo-1.8.8.ebuild

    # The parser isn't very smart and only has basic support for the
    # || ( X Y ) OR logic to do the following:
    # In order to save time needlessly unpacking packages and looking or a
    # cleartext license (which is really a crapshoot), if we have a license
    # like BSD that requires looking for copyright attribution, but we can
    # chose another license like GPL, we do that.

    if not self.ebuild_license_names:
      logging.error("%s: no license found in ebuild. FIXME!", self.fullnamerev)
      # In a bind, you could comment this out. I'm making the output fail to
      # get your attention since this error really should be fixed, but if you
      # comment out the next line, the script will try to find a license inside
      # the source.
      raise PackageLicenseError()

    # This is not invalid, but the parser can't deal with it, so if it ever
    # happens, error out to tell the programmer to do something.
    if "||" in self.ebuild_license_names[1:]:
      raise AssertionError("%s: Can't parse || in the middle of a license: %s"
                           % (self.fullnamerev,
                              ' '.join(self.ebuild_license_names)))

    or_licenses_and_one_is_no_attribution = False
    # We do a quick early pass first so that the longer pass below can
    # run accordingly.
    for license_name in [x for x in self.ebuild_license_names
                         if x not in LICENCES_IGNORE]:
      # Here we have an OR case, and one license that we can use stock, so
      # we remember that in order to be able to skip license attributions if
      # any were in the OR.
      if (self.ebuild_license_names[0] == "||" and
          license_name not in COPYRIGHT_ATTRIBUTION_LICENSES):
        or_licenses_and_one_is_no_attribution = True

    for license_name in [x for x in self.ebuild_license_names
                         if x not in LICENCES_IGNORE]:
      # Licenses like BSD or MIT can't be used as is because they do not contain
      # copyright self. They have to be replaced by copyright file given in the
      # source code, or manually mapped by us in PACKAGE_LICENSES
      if license_name in COPYRIGHT_ATTRIBUTION_LICENSES:
        # To limit needless efforts, if a package is BSD or GPL, we ignore BSD
        # and use GPL to avoid scanning the package, but we can only do this if
        # or_licenses_and_one_is_no_attribution has been set above.
        # This ensures that if we have License: || (BSD3 BSD4), we will
        # look in the source.
        if or_licenses_and_one_is_no_attribution:
          logging.info("%s: ignore license %s because ebuild LICENSES had %s",
                       self.fullnamerev, license_name,
                       ' '.join(self.ebuild_license_names))
        else:
          logging.info("%s: can't use %s, will scan source code for copyright",
                       self.fullnamerev, license_name)
          self.need_copyright_attribution = True
          self.scan_source_for_licenses = True
      else:
        self.license_names.add(license_name)
        # We can't display just 2+ because it only contains text that says to
        # read v2 or v3.
        if license_name == 'GPL-2+':
          self.license_names.add('GPL-2')
        if license_name == 'LGPL-2+':
          self.license_names.add('LGPL-2')

      if license_name in LOOK_IN_SOURCE_LICENSES:
        logging.info("%s: Got %s, will try to find better license in source...",
                     self.fullnamerev, license_name)
        self.scan_source_for_licenses = True

    if self.license_names:
      logging.info('%s: using stock|cust license(s) %s',
                   self.fullnamerev, ','.join(self.license_names))

    # If the license(s) could not be found, or one requires copyright
    # attribution, dig in the source code for license files:
    # For instance:
    # Read licenses from ebuild for net-dialup/ppp-2.4.5-r3: BSD,GPL-2
    # We need get the substitution file for BSD and add it to GPL.
    if self.scan_source_for_licenses:
      self._ExtractLicenses()

    # This shouldn't run, but leaving as sanity check.
    if not self.license_names and not self.license_text_scanned:
      raise AssertionError("Didn't find usable licenses for %s" %
                           self.fullnamerev)


class Licensing(object):
  """Do the actual work of extracting licensing info and outputting html."""

  def __init__(self, board, package_fullnames,
               entry_template_file=ENTRY_TMPL):

    # eg x86-alex
    self.board = board
    # List of stock and custom licenses referenced in ebuilds. Used to
    # print a report. Dict value says which packages use that license.
    self.licenses = {}

    # This keeps track of whether we have an incomplete license file due to
    # package errors during parsing.
    # Any non empty list at the end shows the list of packages that caused
    # errors.
    self.incomplete_packages = []

    self.package_text = {}
    self.entry_template = ReadUnknownEncodedFile(entry_template_file)

    # We need to have a dict for the list of packages objects, index by package
    # fullnamerev, so that when we scan our licenses at the end, and find out
    # some shared licenses are only used by one package, we can access that
    # package object by name, and add the license directly in that object.
    self.packages = {}
    self._package_fullnames = package_fullnames

  @property
  def sorted_licenses(self):
    return sorted(self.licenses.keys(), key=str.lower)

  def LicensedPackages(self, license_name):
    """Return list of packages using a given license."""
    return self.licenses[license_name]

  def ProcessPackages(self):
    """Iterate through all packages provided and gather their licenses.

    GetLicenses will scrape licenses from the code and/or gather stock license
    names. We gather the list of stock and custom ones for later processing.

    Do not call this after adding virtual packages with AddExtraPkg.
    """
    for package_name in self._package_fullnames:
      pkg = PackageInfo(self.board)
      try:
        pkg.GetLicenses(package_name)
        self.packages[package_name] = pkg
      except PackageSkipped as e:
        logging.info(e)
      except PackageLicenseError:
        self.incomplete_packages += [pkg.fullnamerev]

  def AddExtraPkg(self, pkg_data):
    """Allow adding pre-created virtual packages.

    GetLicenses will not work on them, so add them after having run
    ProcessPackages.

    Args:
      pkg_data: array of package data as defined below
    """
    pkg = PackageInfo(self.board)
    pkg.category = pkg_data[0]
    pkg.name = pkg_data[1]
    pkg.version = pkg_data[2]
    pkg.homepages = pkg_data[3]      # this is a list
    pkg.license_names = pkg_data[4]  # this is also a list
    pkg.ebuild_license_names = pkg_data[4]
    self.packages[pkg.fullnamerev] = pkg

  @staticmethod
  def FindLicenseType(license_name):
    """Says if a license is stock Gentoo, custom, or doesn't exist."""

    for directory in STOCK_LICENSE_DIRS:
      path = '%s/%s' % (directory, license_name)
      if os.path.exists(path):
        return "Gentoo Package Stock"

    for directory in CUSTOM_LICENSE_DIRS:
      path = '%s/%s' % (directory, license_name)
      if os.path.exists(path):
        return "Custom"

    raise AssertionError("""
license %s could not be found in %s
If the license in the ebuild is correct,
a) a stock license should be added to portage-stable/licenses :
running `cros_portage_upgrade` inside of the chroot should clone this repo
to /tmp/portage/:
https://chromium.googlesource.com/chromiumos/overlays/portage/+/gentoo
find the new licenses under licenses, and add them to portage-stable/licenses

b) if it's a non gentoo package with a custom license, you can copy that license
to third_party/chromiumos-overlay/licenses/""" %
                           (license_name,
                            '\n'.join(STOCK_LICENSE_DIRS + CUSTOM_LICENSE_DIRS))
                         )

  @staticmethod
  def ReadSharedLicense(license_name):
    """Read and return stock or cust license file specified in an ebuild."""

    license_path = None
    for directory in STOCK_LICENSE_DIRS + CUSTOM_LICENSE_DIRS:
      path = os.path.join(directory, license_name)
      if os.path.exists(path):
        license_path = path
        break

    if license_path:
      return ReadUnknownEncodedFile(license_path, "read license")
    else:
      raise AssertionError("license %s could not be found in %s"
                           % (license_name,
                              '\n'.join(STOCK_LICENSE_DIRS +
                                        CUSTOM_LICENSE_DIRS))
                          )

  @staticmethod
  def EvaluateTemplate(template, env):
    """Expand a template with vars like {{foo}} using a dict of expansions."""
    # TODO switch to stock python templates.
    for key, val in env.iteritems():
      template = template.replace('{{%s}}' % key, val)
    return template

  def _GeneratePackageLicenseText(self, package):
    """Concatenate all licenses related to a package.

    This means a combination of ebuild shared licenses and licenses read from
    the package source tree, if any.

    Args:
      package: PackageInfo object

    Raises:
      AssertionError: on runtime errors
    """

    license_text = []
    for license_text_scanned in package.license_text_scanned:
      license_text.append(license_text_scanned)
      license_text.append('%s\n' % ('-=' * 40))

    license_pointers = []
    # sln: shared license name.
    for sln in package.license_names:
      # Says whether it's a stock gentoo or custom license.
      license_type = self.FindLicenseType(sln)
      license_pointers.append(
          "<li><a href='#%s'>%s License %s</a></li>" % (
              sln, license_type, sln))

    # This should get caught earlier, but one extra check.
    if not (license_text + license_pointers):
      raise AssertionError('Ended up with no license_text')

    env = {
        'name': "%s-%s" % (package.name, package.version),
        'url': cgi.escape(package.homepages[0]) if package.homepages else '',
        'licenses_txt': cgi.escape('\n'.join(license_text)) or '',
        'licenses_ptr': '\n'.join(license_pointers) or '',
    }
    self.package_text[package] = self.EvaluateTemplate(self.entry_template, env)

  def GenerateHTMLLicenseOutput(self, output_file,
                                output_template=TMPL,
                                license_template=SHARED_LICENSE_TMPL):
    """Generate the combined html license file used in ChromeOS.

    Args:
      output_file: resulting HTML license output.
      output_template: template for the entire HTML file.
      license_template: template for shared license entries.
    """
    sorted_license_txt = []

    # Keep track of which licenses are used by which packages.
    for pkg in self.packages.values():
      for sln in pkg.license_names:
        self.licenses.setdefault(sln, []).append(pkg.fullnamerev)

    # Find licenses only used once, and roll them in the package that uses them.
    # We use keys() because licenses is modified in the loop, so we can't use
    # an iterator.
    for sln in self.licenses.keys():
      if len(self.licenses[sln]) == 1:
        pkg_fullnamerev = self.licenses[sln][0]
        logging.info("Collapsing shared license %s into single use license "
                     "(only used by %s)", sln, pkg_fullnamerev)
        license_type = self.FindLicenseType(sln)
        license_txt = self.ReadSharedLicense(sln)
        single_license = "%s License %s:\n\n%s" % (license_type, sln,
                                                   license_txt)
        pkg = self.packages[pkg_fullnamerev]
        pkg.license_text_scanned.append(single_license)
        pkg.license_names.remove(sln)
        del self.licenses[sln]

    for pkg in sorted(self.packages.values(),
                      key=lambda x: (x.name.lower(), x.version, x.revision)):
      self._GeneratePackageLicenseText(pkg)
      sorted_license_txt += [self.package_text[pkg]]

    # Now generate the bottom of the page that will contain all the shared
    # licenses and a list of who is pointing to them.
    license_template = ReadUnknownEncodedFile(license_template)

    licenses_txt = []
    for license_name in self.sorted_licenses:
      env = {
          'license_name': license_name,
          'license': cgi.escape(self.ReadSharedLicense(license_name)),
          'license_type': self.FindLicenseType(license_name),
          'license_packages': ' '.join(self.LicensedPackages(license_name)),
      }
      licenses_txt += [self.EvaluateTemplate(license_template, env)]

    file_template = ReadUnknownEncodedFile(output_template)
    env = {
        'entries': '\n'.join(sorted_license_txt),
        'licenses': '\n'.join(licenses_txt),
    }
    osutils.WriteFile(output_file,
                      self.EvaluateTemplate(file_template, env).encode('UTF-8'))


def ListInstalledPackages(board):
  """Return a list of all packages installed for a particular board."""

  # Please leave the following as documentation of what options are available
  # and why we chose the current one.

  # The following returns all packages that were part of the build tree
  # (many get built or used during the build, but do not get shipped).
  # Note that it also contains packages that are in the build as
  # defined by build_packages but not part of the image we ship.
  # args = "/usr/local/bin/equery-%s list '*'" % board

  # This is better because it only lists packages that are in the
  # chromeos-base/chromeos dependency chain and ignores packages that
  # just happen to be available in your build tree.
  # args = "emerge-%s --with-bdeps=y --usepkgonly --emptytree \
  #       --pretend chromeos-base/chromeos | \
  #       sed -n -E -e 's/^\[[^]]*\] ([^ ]*) .*$/\\1/p'" % board

  # Another option which we've decided not to use, is bdeps=n. This
  # outputs just the packages we ship, but does not list packages
  # that were used to build them, including a package like flex which
  # generates a .a that is included and shipped in ChromeOS.
  # We've decided to credit build packages, even if we're not legally
  # required to (it's always nice to do), and that way we get corner
  # case packages like flex.
  #
  # return cros_build_lib.RunCommand(args, print_cmd=debug, shell=True,
  #                                 redirect_stdout=True).output.splitlines()

  args = ["emerge-%s" % board, "--with-bdeps=y", "--usepkgonly", "--emptytree",
          "--pretend", "--color=n", "chromeos-base/chromeos"]
  emerge = cros_build_lib.RunCommand(args, print_cmd=debug,
                                     redirect_stdout=True).output.splitlines()

  packages = []
  # [binary   R    ] x11-libs/libva-1.1.1 to /build/x86-alex/
  pkg_rgx = re.compile(r'\[[^]]+\] (.+) to /build/.*')
  for line in emerge:
    match = pkg_rgx.search(line)
    if match:
      packages.append(match.group(1))

  return packages


def _HandleIllegalXMLChars(text):
  """Handles illegal XML Characters.

  XML 1.0 acceptable character range:
  Char ::= #x9 | #xA | #xD | [#x20-#xD7FF] | [#xE000-#xFFFD] | \
           [#x10000-#x10FFFF]

  This function finds all illegal characters in the text and filters
  out all whitelisted characters (e.g. ^L).

  Args:
    text: text to examine.

  Returns:
    Filtered |text| and a list of non-whitelisted illegal characters found.
  """
  whitelist_re = re.compile(u'[\x0c]')
  text = whitelist_re.sub('', text)
  # illegal_chars_re includes all illegal characters (whitelisted or
  # not), so we can expand the whitelist without modifying this line.
  illegal_chars_re = re.compile(
      u'[\x00-\x08\x0b\x0c\x0e-\x1F\uD800-\uDFFF\uFFFE\uFFFF]')
  return (text, illegal_chars_re.findall(text))


def ReadUnknownEncodedFile(file_path, logging_text=None):
  """Read a file of unknown encoding (UTF-8 or latin) by trying in sequence.

  Args:
    file_path: what to read.
    logging_text: what to display for logging depending on file read.

  Returns:
    File content, possibly converted from latin1 to UTF-8.

  Raises:
    Assertion error: if non-whitelisted illegal XML characters
      are found in the file.
  """

  try:
    with codecs.open(file_path, encoding="utf-8") as c:
      file_txt = c.read()
      if logging_text:
        logging.info("%s %s (UTF-8)", logging_text, file_path)
  except UnicodeDecodeError:
    with codecs.open(file_path, encoding="latin1") as c:
      file_txt = c.read()
      if logging_text:
        logging.info("%s %s (latin1)", logging_text, file_path)

  file_txt, char_list = _HandleIllegalXMLChars(file_txt)

  if char_list:
    raise ValueError('Illegal XML characters %s found in %s.' %
                     (char_list, file_path))

  return file_txt


def main(args):
  # pylint: disable=W0603
  global SKIPPED_PACKAGES
  global debug
  # pylint: enable=W0603

  parser = commandline.ArgumentParser(usage=__doc__)
  parser.add_argument("-b", "--board", default=cros_build_lib.GetDefaultBoard(),
                      help="which board to run for, like x86-alex")
  parser.add_argument("-p", "--package", action="append", default=[],
                      help="check the license of the package, e.g.,"
                      "dev-libs/libatomic_ops-7.2d")
  parser.add_argument("-o", "--output", type="path",
                      help="which html file to create with output")
  opts = parser.parse_args(args)
  debug = opts.debug
  board, output_file = opts.board, opts.output
  logging.info("Using board %s.", board)

  packages_mode = bool(opts.package)

  if not output_file and not packages_mode:
    logging.warning('No output file is specified.')

  builddir = os.path.join(cros_build_lib.GetSysroot(board=board),
                          'tmp', 'portage')
  if not os.path.exists(builddir):
    raise AssertionError(
        "FATAL: %s missing.\n"
        "Did you give the right board and build that tree?" % builddir)

  # We have a hardcoded list of skipped packages for various reasons,
  # but we also exclude any google platform package from needing a
  # license since they are covered by the top license in the tree.
  # TODO: find a better way to deal with this because running
  # cros_workon for individual packages (with -p) is too costly.
  cmd = "cros_workon info --all --host | grep src/platform/ | awk '{print $1}'"
  packages = cros_build_lib.RunCommand(cmd, shell=True, print_cmd=debug,
                                       redirect_stdout=True).output.splitlines()
  SKIPPED_PACKAGES += packages

  if packages_mode:
    packages = opts.package
  else:
    packages = ListInstalledPackages(board)
  if not packages:
    raise AssertionError('FATAL: Could not get any packages for board %s' %
                         board)
  logging.debug("Package list to work through:")
  logging.debug('\n'.join(packages))
  logging.debug("Will skip these packages:")
  logging.debug('\n'.join(SKIPPED_PACKAGES))
  licensing = Licensing(board, packages)
  licensing.ProcessPackages()
  if not packages_mode:
    # We add 2 virtual packages as well as 2 boot packages that are included
    # with some hardware, but not in the image or package list.
    for extra_pkg in [
        ['x11-base', 'X.Org', '1.9.3', ['http://www.x.org/'], ['X']],
        ['sys-kernel', 'Linux', '2.6', ['http://www.kernel.org/'], ['GPL-2']],
        ['sys-boot', 'u-boot', '2013.06', ['http://www.denx.de/wiki/U-Boot'],
         ['GPL-2+']],
        ['sys-boot', 'coreboot', '2013.04', ['http://www.coreboot.org/'],
         ['GPL-2']],
    ]:
      licensing.AddExtraPkg(extra_pkg)

  if output_file:
    licensing.GenerateHTMLLicenseOutput(output_file)

  if licensing.incomplete_packages:
    raise AssertionError("""
DO NOT USE OUTPUT!!!
Some packages are missing due to errors, please look at errors generated during
this run.
List of packages with errors:
%s
  """ % '\n'.join(licensing.incomplete_packages))