tools/fonts/fontchain_lint.py

#!/usr/bin/env python

import collections
import copy
import glob
import itertools
from os import path
import sys
from xml.etree import ElementTree

from fontTools import ttLib

EMOJI_VS = 0xFE0F

LANG_TO_SCRIPT = {
    'as': 'Beng',
    'bn': 'Beng',
    'cy': 'Latn',
    'da': 'Latn',
    'de': 'Latn',
    'en': 'Latn',
    'es': 'Latn',
    'et': 'Latn',
    'eu': 'Latn',
    'fr': 'Latn',
    'ga': 'Latn',
    'gu': 'Gujr',
    'hi': 'Deva',
    'hr': 'Latn',
    'hu': 'Latn',
    'hy': 'Armn',
    'ja': 'Jpan',
    'kn': 'Knda',
    'ko': 'Kore',
    'ml': 'Mlym',
    'mn': 'Cyrl',
    'mr': 'Deva',
    'nb': 'Latn',
    'nn': 'Latn',
    'or': 'Orya',
    'pa': 'Guru',
    'pt': 'Latn',
    'sl': 'Latn',
    'ta': 'Taml',
    'te': 'Telu',
    'tk': 'Latn',
}

def lang_to_script(lang_code):
    lang = lang_code.lower()
    while lang not in LANG_TO_SCRIPT:
        hyphen_idx = lang.rfind('-')
        assert hyphen_idx != -1, (
            'We do not know what script the "%s" language is written in.'
            % lang_code)
        assumed_script = lang[hyphen_idx+1:]
        if len(assumed_script) == 4 and assumed_script.isalpha():
            # This is actually the script
            return assumed_script.title()
        lang = lang[:hyphen_idx]
    return LANG_TO_SCRIPT[lang]


def printable(inp):
    if type(inp) is set:  # set of character sequences
        return '{' + ', '.join([printable(seq) for seq in inp]) + '}'
    if type(inp) is tuple:  # character sequence
        return '<' + (', '.join([printable(ch) for ch in inp])) + '>'
    else:  # single character
        return 'U+%04X' % inp


def open_font(font):
    font_file, index = font
    font_path = path.join(_fonts_dir, font_file)
    if index is not None:
        return ttLib.TTFont(font_path, fontNumber=index)
    else:
        return ttLib.TTFont(font_path)


def get_best_cmap(font):
    ttfont = open_font(font)
    all_unicode_cmap = None
    bmp_cmap = None
    for cmap in ttfont['cmap'].tables:
        specifier = (cmap.format, cmap.platformID, cmap.platEncID)
        if specifier == (4, 3, 1):
            assert bmp_cmap is None, 'More than one BMP cmap in %s' % (font, )
            bmp_cmap = cmap
        elif specifier == (12, 3, 10):
            assert all_unicode_cmap is None, (
                'More than one UCS-4 cmap in %s' % (font, ))
            all_unicode_cmap = cmap

    return all_unicode_cmap.cmap if all_unicode_cmap else bmp_cmap.cmap


def get_variation_sequences_cmap(font):
    ttfont = open_font(font)
    vs_cmap = None
    for cmap in ttfont['cmap'].tables:
        specifier = (cmap.format, cmap.platformID, cmap.platEncID)
        if specifier == (14, 0, 5):
            assert vs_cmap is None, 'More than one VS cmap in %s' % (font, )
            vs_cmap = cmap
    return vs_cmap


def get_emoji_map(font):
    # Add normal characters
    emoji_map = copy.copy(get_best_cmap(font))
    reverse_cmap = {glyph: code for code, glyph in emoji_map.items()}

    # Add variation sequences
    vs_dict = get_variation_sequences_cmap(font).uvsDict
    for vs in vs_dict:
        for base, glyph in vs_dict[vs]:
            if glyph is None:
                emoji_map[(base, vs)] = emoji_map[base]
            else:
                emoji_map[(base, vs)] = glyph

    # Add GSUB rules
    ttfont = open_font(font)
    for lookup in ttfont['GSUB'].table.LookupList.Lookup:
        assert lookup.LookupType == 4, 'We only understand type 4 lookups'
        for subtable in lookup.SubTable:
            ligatures = subtable.ligatures
            for first_glyph in ligatures:
                for ligature in ligatures[first_glyph]:
                    sequence = [first_glyph] + ligature.Component
                    sequence = [reverse_cmap[glyph] for glyph in sequence]
                    sequence = tuple(sequence)
                    # Make sure no starting subsequence of 'sequence' has been
                    # seen before.
                    for sub_len in range(2, len(sequence)+1):
                        subsequence = sequence[:sub_len]
                        assert subsequence not in emoji_map
                    emoji_map[sequence] = ligature.LigGlyph

    return emoji_map


def assert_font_supports_any_of_chars(font, chars):
    best_cmap = get_best_cmap(font)
    for char in chars:
        if char in best_cmap:
            return
    sys.exit('None of characters in %s were found in %s' % (chars, font))


def assert_font_supports_all_of_chars(font, chars):
    best_cmap = get_best_cmap(font)
    for char in chars:
        assert char in best_cmap, (
            'U+%04X was not found in %s' % (char, font))


def assert_font_supports_none_of_chars(font, chars):
    best_cmap = get_best_cmap(font)
    for char in chars:
        assert char not in best_cmap, (
            'U+%04X was found in %s' % (char, font))


def assert_font_supports_all_sequences(font, sequences):
    vs_dict = get_variation_sequences_cmap(font).uvsDict
    for base, vs in sorted(sequences):
        assert vs in vs_dict and (base, None) in vs_dict[vs], (
            '<U+%04X, U+%04X> was not found in %s' % (base, vs, font))


def check_hyphens(hyphens_dir):
    # Find all the scripts that need automatic hyphenation
    scripts = set()
    for hyb_file in glob.iglob(path.join(hyphens_dir, '*.hyb')):
        hyb_file = path.basename(hyb_file)
        assert hyb_file.startswith('hyph-'), (
            'Unknown hyphenation file %s' % hyb_file)
        lang_code = hyb_file[hyb_file.index('-')+1:hyb_file.index('.')]
        scripts.add(lang_to_script(lang_code))

    HYPHENS = {0x002D, 0x2010}
    for script in scripts:
        fonts = _script_to_font_map[script]
        assert fonts, 'No fonts found for the "%s" script' % script
        for font in fonts:
            assert_font_supports_any_of_chars(font, HYPHENS)


class FontRecord(object):
    def __init__(self, name, scripts, variant, weight, style, font):
        self.name = name
        self.scripts = scripts
        self.variant = variant
        self.weight = weight
        self.style = style
        self.font = font


def parse_fonts_xml(fonts_xml_path):
    global _script_to_font_map, _fallback_chain
    _script_to_font_map = collections.defaultdict(set)
    _fallback_chain = []
    tree = ElementTree.parse(fonts_xml_path)
    for family in tree.findall('family'):
        name = family.get('name')
        variant = family.get('variant')
        langs = family.get('lang')
        if name:
            assert variant is None, (
                'No variant expected for LGC font %s.' % name)
            assert langs is None, (
                'No language expected for LGC fonts %s.' % name)
        else:
            assert variant in {None, 'elegant', 'compact'}, (
                'Unexpected value for variant: %s' % variant)

        if langs:
            langs = langs.split()
            scripts = {lang_to_script(lang) for lang in langs}
        else:
            scripts = set()

        for child in family:
            assert child.tag == 'font', (
                'Unknown tag <%s>' % child.tag)
            font_file = child.text
            weight = int(child.get('weight'))
            assert weight % 100 == 0, (
                'Font weight "%d" is not a multiple of 100.' % weight)

            style = child.get('style')
            assert style in {'normal', 'italic'}, (
                'Unknown style "%s"' % style)

            index = child.get('index')
            if index:
                index = int(index)

            _fallback_chain.append(FontRecord(
                name,
                frozenset(scripts),
                variant,
                weight,
                style,
                (font_file, index)))

            if name: # non-empty names are used for default LGC fonts
                map_scripts = {'Latn', 'Grek', 'Cyrl'}
            else:
                map_scripts = scripts
            for script in map_scripts:
                _script_to_font_map[script].add((font_file, index))


def check_emoji_coverage(all_emoji, equivalent_emoji):
  emoji_font = get_emoji_font()
  check_emoji_font_coverage(emoji_font, all_emoji, equivalent_emoji)


def get_emoji_font():
    emoji_fonts = [
        record.font for record in _fallback_chain
        if 'Zsye' in record.scripts]
    assert len(emoji_fonts) == 1, 'There are %d emoji fonts.' % len(emoji_fonts)
    return emoji_fonts[0]


def check_emoji_font_coverage(emoji_font, all_emoji, equivalent_emoji):
    coverage = get_emoji_map(emoji_font)
    for sequence in all_emoji:
        assert sequence in coverage, (
            '%s is not supported in the emoji font.' % printable(sequence))

    # disable temporarily - we cover more than this
    """
    for sequence in coverage:
        if sequence in {0x0000, 0x000D, 0x0020}:
            # The font needs to support a few extra characters, which is OK
            continue
        assert sequence in all_emoji, (
            'Emoji font should not support %s.' % printable(sequence))
    """

    for first, second in sorted(equivalent_emoji.items()):
        assert coverage[first] == coverage[second], (
            '%s and %s should map to the same glyph.' % (
                printable(first),
                printable(second)))

    # disable temporarily - some equivalent sequences we don't even know about
    """
    for glyph in set(coverage.values()):
        maps_to_glyph = [seq for seq in coverage if coverage[seq] == glyph]
        if len(maps_to_glyph) > 1:
            # There are more than one sequences mapping to the same glyph. We
            # need to make sure they were expected to be equivalent.
            equivalent_seqs = set()
            for seq in maps_to_glyph:
                equivalent_seq = seq
                while equivalent_seq in equivalent_emoji:
                    equivalent_seq = equivalent_emoji[equivalent_seq]
                equivalent_seqs.add(equivalent_seq)
            assert len(equivalent_seqs) == 1, (
                'The sequences %s should not result in the same glyph %s' % (
                    printable(equivalent_seqs),
                    glyph))
    """

def check_emoji_defaults(default_emoji):
    missing_text_chars = _emoji_properties['Emoji'] - default_emoji
    emoji_font_seen = False
    for record in _fallback_chain:
        if 'Zsye' in record.scripts:
            emoji_font_seen = True
            # No need to check the emoji font
            continue
        # For later fonts, we only check them if they have a script
        # defined, since the defined script may get them to a higher
        # score even if they appear after the emoji font.
        if emoji_font_seen and not record.scripts:
            continue

        # Check default emoji-style characters
        assert_font_supports_none_of_chars(record.font, sorted(default_emoji))

        # Mark default text-style characters appearing in fonts above the emoji
        # font as seen
        if not emoji_font_seen:
            missing_text_chars -= set(get_best_cmap(record.font))

    # Noto does not have monochrome glyphs for Unicode 7.0 wingdings and
    # webdings yet.
    missing_text_chars -= _chars_by_age['7.0']
    # TODO: Remove these after b/26113320 is fixed
    missing_text_chars -= {
        0x263A, # WHITE SMILING FACE
        0x270C, # VICTORY HAND
        0x2744, # SNOWFLAKE
        0x2764, # HEAVY BLACK HEART
    }
    assert missing_text_chars == set(), (
        'Text style version of some emoji characters are missing: ' + repr(missing_text_chars))


# Setting reverse to true returns a dictionary that maps the values to sets of
# characters, useful for some binary properties. Otherwise, we get a
# dictionary that maps characters to the property values, assuming there's only
# one property in the file.
def parse_unicode_datafile(file_path, reverse=False):
    if reverse:
        output_dict = collections.defaultdict(set)
    else:
        output_dict = {}
    with open(file_path) as datafile:
        for line in datafile:
            if '#' in line:
                line = line[:line.index('#')]
            line = line.strip()
            if not line:
                continue

            chars, prop = line.split(';')
            chars = chars.strip()
            prop = prop.strip()

            if ' ' in chars:  # character sequence
                sequence = [int(ch, 16) for ch in chars.split(' ')]
                additions = [tuple(sequence)]
            elif '..' in chars:  # character range
                char_start, char_end = chars.split('..')
                char_start = int(char_start, 16)
                char_end = int(char_end, 16)
                additions = xrange(char_start, char_end+1)
            else:  # singe character
                additions = [int(chars, 16)]
            if reverse:
                output_dict[prop].update(additions)
            else:
                for addition in additions:
                    assert addition not in output_dict
                    output_dict[addition] = prop
    return output_dict


def parse_standardized_variants(file_path):
    emoji_set = set()
    text_set = set()
    with open(file_path) as datafile:
        for line in datafile:
            if '#' in line:
                line = line[:line.index('#')]
            line = line.strip()
            if not line:
                continue
            sequence, description, _ = line.split(';')
            sequence = sequence.strip().split(' ')
            base = int(sequence[0], 16)
            vs = int(sequence[1], 16)
            description = description.strip()
            if description == 'text style':
                text_set.add((base, vs))
            elif description == 'emoji style':
                emoji_set.add((base, vs))
    return text_set, emoji_set


def parse_ucd(ucd_path):
    global _emoji_properties, _chars_by_age
    global _text_variation_sequences, _emoji_variation_sequences
    global _emoji_sequences, _emoji_zwj_sequences
    _emoji_properties = parse_unicode_datafile(
        path.join(ucd_path, 'emoji-data.txt'), reverse=True)
    _chars_by_age = parse_unicode_datafile(
        path.join(ucd_path, 'DerivedAge.txt'), reverse=True)
    sequences = parse_standardized_variants(
        path.join(ucd_path, 'StandardizedVariants.txt'))
    _text_variation_sequences, _emoji_variation_sequences = sequences
    _emoji_sequences = parse_unicode_datafile(
        path.join(ucd_path, 'emoji-sequences.txt'))
    _emoji_zwj_sequences = parse_unicode_datafile(
        path.join(ucd_path, 'emoji-zwj-sequences.txt'))

    # filter modern pentathlon, as it seems likely to be removed from final spec
    # also filter rifle
    def is_excluded(n):
        return n in [0x1f93b, 0x1f946]

    def contains_excluded(t):
        if type(t) == int:
            return is_excluded(t)
        return any(is_excluded(cp) for cp in t)

    # filter modern pentathlon, as it seems likely to be removed from final spec
    _emoji_properties['Emoji'] = set(
        t for t in _emoji_properties['Emoji'] if not contains_excluded(t))
    _emoji_sequences = dict(
        (t, v) for (t, v) in _emoji_sequences.items() if not contains_excluded(t))

    # add in UN flag
    UN_seq = flag_sequence('UN')
    _emoji_sequences[UN_seq] = 'Emoji_Flag_Sequence'


def flag_sequence(territory_code):
    return tuple(0x1F1E6 + ord(ch) - ord('A') for ch in territory_code)


UNSUPPORTED_FLAGS = frozenset({
    flag_sequence('BL'), flag_sequence('BQ'), flag_sequence('DG'),
    flag_sequence('EA'), flag_sequence('EH'), flag_sequence('FK'),
    flag_sequence('GF'), flag_sequence('GP'), flag_sequence('GS'),
    flag_sequence('MF'), flag_sequence('MQ'), flag_sequence('NC'),
    flag_sequence('PM'), flag_sequence('RE'), flag_sequence('TF'),
    flag_sequence('WF'), flag_sequence('XK'), flag_sequence('YT'),
})

EQUIVALENT_FLAGS = {
    flag_sequence('BV'): flag_sequence('NO'),
    flag_sequence('CP'): flag_sequence('FR'),
    flag_sequence('HM'): flag_sequence('AU'),
    flag_sequence('SJ'): flag_sequence('NO'),
    flag_sequence('UM'): flag_sequence('US'),
}

COMBINING_KEYCAP = 0x20E3

LEGACY_ANDROID_EMOJI = {
    0xFE4E5: flag_sequence('JP'),
    0xFE4E6: flag_sequence('US'),
    0xFE4E7: flag_sequence('FR'),
    0xFE4E8: flag_sequence('DE'),
    0xFE4E9: flag_sequence('IT'),
    0xFE4EA: flag_sequence('GB'),
    0xFE4EB: flag_sequence('ES'),
    0xFE4EC: flag_sequence('RU'),
    0xFE4ED: flag_sequence('CN'),
    0xFE4EE: flag_sequence('KR'),
    0xFE82C: (ord('#'), COMBINING_KEYCAP),
    0xFE82E: (ord('1'), COMBINING_KEYCAP),
    0xFE82F: (ord('2'), COMBINING_KEYCAP),
    0xFE830: (ord('3'), COMBINING_KEYCAP),
    0xFE831: (ord('4'), COMBINING_KEYCAP),
    0xFE832: (ord('5'), COMBINING_KEYCAP),
    0xFE833: (ord('6'), COMBINING_KEYCAP),
    0xFE834: (ord('7'), COMBINING_KEYCAP),
    0xFE835: (ord('8'), COMBINING_KEYCAP),
    0xFE836: (ord('9'), COMBINING_KEYCAP),
    0xFE837: (ord('0'), COMBINING_KEYCAP),
}

ZWJ_IDENTICALS = {
    # KISS
    (0x1F469, 0x200D, 0x2764, 0x200D, 0x1F48B, 0x200D, 0x1F468): 0x1F48F,
    # COUPLE WITH HEART
    (0x1F469, 0x200D, 0x2764, 0x200D, 0x1F468): 0x1F491,
    # FAMILY
    (0x1F468, 0x200D, 0x1F469, 0x200D, 0x1F466): 0x1F46A,
}


def is_fitzpatrick_modifier(cp):
  return 0x1f3fb <= cp <= 0x1f3ff


def compute_expected_emoji():
    equivalent_emoji = {}
    sequence_pieces = set()
    all_sequences = set()
    all_sequences.update(_emoji_variation_sequences)

    for sequence in _emoji_sequences.keys():
        sequence = tuple(ch for ch in sequence if ch != EMOJI_VS)
        all_sequences.add(sequence)
        sequence_pieces.update(sequence)

    for sequence in _emoji_zwj_sequences.keys():
        sequence = tuple(ch for ch in sequence if ch != EMOJI_VS)
        all_sequences.add(sequence)
        sequence_pieces.update(sequence)
        # Add reverse of all emoji ZWJ sequences, which are added to the fonts
        # as a workaround to get the sequences work in RTL text.
        reversed_seq = list(reversed(sequence))
        # if there are fitzpatrick modifiers in the sequence, keep them after
        # the emoji they modify
        for i in xrange(1, len(reversed_seq)):
          if is_fitzpatrick_modifier(reversed_seq[i - 1]):
            tmp = reversed_seq[i]
            reversed_seq[i] = reversed_seq[i-1]
            reversed_seq[i-1] = tmp
        reversed_seq = tuple(reversed_seq)
        all_sequences.add(reversed_seq)
        equivalent_emoji[reversed_seq] = sequence

    # Add all two-letter flag sequences, as even the unsupported ones should
    # resolve to a flag tofu.
    all_letters = [chr(code) for code in range(ord('A'), ord('Z')+1)]
    all_two_letter_codes = itertools.product(all_letters, repeat=2)
    all_flags = {flag_sequence(code) for code in all_two_letter_codes}
    all_sequences.update(all_flags)
    tofu_flags = UNSUPPORTED_FLAGS | (all_flags - set(_emoji_sequences.keys()))

    all_emoji = (
        _emoji_properties['Emoji'] |
        all_sequences |
        sequence_pieces |
        set(LEGACY_ANDROID_EMOJI.keys()))
    default_emoji = (
        _emoji_properties['Emoji_Presentation'] |
        all_sequences |
        set(LEGACY_ANDROID_EMOJI.keys()))

    first_tofu_flag = sorted(tofu_flags)[0]
    for flag in tofu_flags:
        if flag != first_tofu_flag:
            equivalent_emoji[flag] = first_tofu_flag
    equivalent_emoji.update(EQUIVALENT_FLAGS)
    equivalent_emoji.update(LEGACY_ANDROID_EMOJI)
    equivalent_emoji.update(ZWJ_IDENTICALS)
    for seq in _emoji_variation_sequences:
        equivalent_emoji[seq] = seq[0]

    return all_emoji, default_emoji, equivalent_emoji


def main():
    global _fonts_dir
    target_out = sys.argv[1]
    _fonts_dir = path.join(target_out, 'fonts')

    fonts_xml_path = path.join(target_out, 'etc', 'fonts.xml')
    parse_fonts_xml(fonts_xml_path)

    hyphens_dir = path.join(target_out, 'usr', 'hyphen-data')
    check_hyphens(hyphens_dir)

    check_emoji = sys.argv[2]
    if check_emoji == 'true':
        ucd_path = sys.argv[3]
        parse_ucd(ucd_path)
        all_emoji, default_emoji, equivalent_emoji = compute_expected_emoji()
        check_emoji_coverage(all_emoji, equivalent_emoji)
        check_emoji_defaults(default_emoji)


if __name__ == '__main__':
    main()