colrv1_postproc.py

"""
Post-nanoemoji processing of the Noto COLRv1 Emoji files.

Adds additional sequences to properly support Safari, corrects 'name', etc.

For now substantially based on copying from a correct bitmap build.
"""
from absl import app
import functools
from fontTools.feaLib.builder import addOpenTypeFeaturesFromString
from fontTools import ttLib
from fontTools.ttLib.tables import _g_l_y_f as glyf
from fontTools.ttLib.tables import otTables as ot
import map_pua_emoji
from nototools import add_vs_cmap
from nototools import font_data
from nototools import unicode_data
from pathlib import Path
import re

from colrv1_add_soft_light_to_flags import add_soft_light_to_flags


_CBDT_FILE = Path("fonts/NotoColorEmoji.ttf")
_COLR_FILES = {
    Path("fonts/Noto-COLRv1-noflags.ttf"),
    Path("fonts/Noto-COLRv1.ttf"),
}


def _is_colrv1(font):
    return "COLR" in font and font["COLR"].version == 1


def _is_cbdt(font):
    return "CBDT" in font


def _set_name(name_table, nameID):
    name_table.getName(value, nameID, 3, 1, 0x409)


def _set_name(name_table, nameID, value):
    name_table.setName(value, nameID, 3, 1, 0x409)


def _copy_names(colr_font, cbdt_font):
    colr_font["name"] = cbdt_font["name"]
    name_table = colr_font["name"]
    assert all(
        (n.platformID, n.platEncID, n.langID) == (3, 1, 0x409) for n in name_table.names
    ), "Should only have names Android uses"

    # Amendments
    _set_name(name_table, 10, "Color emoji font using COLRv1.")
    _set_name(name_table, 11, "https://github.com/googlefonts/noto-emoji")
    _set_name(name_table, 12, "https://github.com/googlefonts/noto-emoji")


# CBDT build step: @$(VS_ADDER) -vs 2640 2642 2695 --dstdir '.' -o "$@-with-pua-varsel" "$@-with-pua"
def _add_vs_cmap(colr_font):
    emoji_variants = unicode_data.get_unicode_emoji_variants() | {
        0x2640,
        0x2642,
        0x2695,
    }
    add_vs_cmap.modify_font("COLRv1 Emoji", colr_font, "emoji", emoji_variants)


def _is_variation_selector_cmap_table(table):
    assert table.format in {4, 12, 14}
    return table.format == 14


def _lookup_in_cmap(colr_font, codepoint):
    result = set()
    for table in colr_font["cmap"].tables:
        if _is_variation_selector_cmap_table(table):
            continue
        assert codepoint in table.cmap
        result.add(table.cmap[codepoint])
    assert len(result) == 1, f"Ambiguous mapping for {codepoint}: {result}"
    return next(iter(result))


def _add_cmap_entries(colr_font, codepoint, glyph_name):
    for table in colr_font["cmap"].tables:
        if _is_variation_selector_cmap_table(table):
            continue
        if not _is_bmp(codepoint) and table.format == 4:
            continue
        table.cmap[codepoint] = glyph_name
        #print(f"Map 0x{codepoint:04x} to {glyph_name}, format {table.format}")


FLAG_TAGS = set(range(0xE0030, 0xE0039 + 1)) | set(range(0xE0061, 0xE007A + 1))
CANCEL_TAG = 0xE007F


def _map_missing_flag_tag_chars_to_empty_glyphs(colr_font):
    # Add all tag characters used in flags + cancel tag
    tag_cps = FLAG_TAGS | {CANCEL_TAG}

    # Anything already cmap'd is fine
    tag_cps -= set(_Cmap(colr_font).keys())

    # CBDT maps these to blank glyphs
    glyf_table = colr_font["glyf"]
    hmtx_table = colr_font["hmtx"]
    glyph_order_size = len(glyf_table.glyphOrder)
    for cp in tag_cps:
        #print(f"Map 0x{cp:04x} to a blank glyf")
        glyph_name = f"u{cp:04X}"
        assert glyph_name not in glyf_table, f"{glyph_name} already in glyf"
        assert glyph_name not in hmtx_table.metrics, f"{glyph_name} already in hmtx"
        glyf_table[glyph_name] = glyf.Glyph()
        hmtx_table[glyph_name] = (0, 0)

        _add_cmap_entries(colr_font, cp, glyph_name)


def _is_bmp(cp):
    return cp in range(0x0000, 0xFFFF + 1)


def _ligaset_for_glyph(lookup_list, glyph_name):
    for lookup in lookup_list.Lookup:
        if lookup.LookupType != 4:
            continue
        for liga_set in lookup.SubTable:
            if glyph_name in liga_set.ligatures:
                return liga_set.ligatures[glyph_name]
    return None


def _Cmap(ttfont):
    def _Reducer(acc, u):
        acc.update(u)
        return acc

    unicode_cmaps = (t.cmap for t in ttfont["cmap"].tables if t.isUnicode())
    return functools.reduce(_Reducer, unicode_cmaps, {})


def _add_vertical_layout_tables(cbdt_font, colr_font):
    upem_scale = colr_font["head"].unitsPerEm / cbdt_font["head"].unitsPerEm

    vhea = colr_font["vhea"] = ttLib.newTable("vhea")
    vhea.tableVersion = 0x00010000
    vhea.ascent = round(cbdt_font["vhea"].ascent * upem_scale)
    vhea.descent = round(cbdt_font["vhea"].descent * upem_scale)
    vhea.lineGap = 0
    # most of the stuff below is recalculated by the compiler, but still needs to be
    # initialized... ¯\_(ツ)_/¯
    vhea.advanceHeightMax = 0
    vhea.minTopSideBearing = 0
    vhea.minBottomSideBearing = 0
    vhea.yMaxExtent = 0
    vhea.caretSlopeRise = 0
    vhea.caretSlopeRun = 0
    vhea.caretOffset = 0
    vhea.reserved0 = 0
    vhea.reserved1 = 0
    vhea.reserved2 = 0
    vhea.reserved3 = 0
    vhea.reserved4 = 0
    vhea.metricDataFormat = 0
    vhea.numberOfVMetrics = 0

    # emoji font is monospaced -- except for an odd uni0000 (NULL) glyph which happens
    # to have height=0; but colrv1 font doesn't have that anyway, so I just skip it
    cbdt_heights = set(h for h, _ in cbdt_font["vmtx"].metrics.values() if h != 0)
    assert len(cbdt_heights) == 1, "NotoColorEmoji CBDT font should be monospaced!"
    height = round(cbdt_heights.pop() * upem_scale)
    vmtx = colr_font["vmtx"] = ttLib.newTable("vmtx")
    vmtx.metrics = {}
    for gn in colr_font.getGlyphOrder():
        vmtx.metrics[gn] = height, 0


UNKNOWN_FLAG_PUA = 0xFE82B
BLACK_FLAG = 0x1F3F4
REGIONAL_INDICATORS = set(range(0x1F1E6, 0x1F1FF + 1))


def _add_fallback_subs_for_unknown_flags(colr_font):
    """Add GSUB lookups to replace unsupported flag sequences with the 'unknown flag'.

    In order to locate the unknown flag, the glyph must be mapped to 0xFE82B PUA code;
    the latter is removed from the cmap table after the GSUB has been updated.
    """
    cmap = _Cmap(colr_font)
    unknown_flag = cmap[UNKNOWN_FLAG_PUA]
    black_flag = cmap[BLACK_FLAG]
    cancel_tag = cmap[CANCEL_TAG]
    flag_tags = sorted(cmap[cp] for cp in FLAG_TAGS)
    # in the *-noflags.ttf font there are no region flags thus this list is empty
    regional_indicators = sorted(cmap[cp] for cp in REGIONAL_INDICATORS if cp in cmap)

    classes = f'@FLAG_TAGS = [{" ".join(flag_tags)}];\n'
    if regional_indicators:
        classes += f"""
            @REGIONAL_INDICATORS = [{" ".join(regional_indicators)}];
            @UNKNOWN_FLAG = [{" ".join([unknown_flag] * len(regional_indicators))}];
        """
    lookups = (
        # the first lookup is a dummy that stands for the emoji sequences ligatures
        # from the destination font; we only use it to ensure the lookup indices match.
        # We can't leave it empty otherwise feaLib optimizes it away.
        f"""
        lookup placeholder {{
            sub {unknown_flag} {unknown_flag} by {unknown_flag};
        }} placeholder;
        """
        + "\n".join(
            ["lookup delete_glyph {"]
            + [f"    sub {g} by NULL;" for g in sorted(regional_indicators + flag_tags)]
            + ["} delete_glyph;"]
        )
        + (
            """
            lookup replace_with_unknown_flag {
                sub @REGIONAL_INDICATORS by @UNKNOWN_FLAG;
            } replace_with_unknown_flag;
            """
            if regional_indicators
            else "\n"
        )
    )
    features = (
        "languagesystem DFLT dflt;\n"
        + classes
        + lookups
        + "feature ccmp {"
        + f"""
            lookup placeholder;
            sub {black_flag} @FLAG_TAGS' lookup delete_glyph;
            sub {black_flag} {cancel_tag} by {unknown_flag};
        """
        + (
            """
            sub @REGIONAL_INDICATORS' lookup replace_with_unknown_flag
                @REGIONAL_INDICATORS' lookup delete_glyph;
            """
            if regional_indicators
            else ""
        )
        + "} ccmp;"
    )
    # feaLib always builds a new GSUB table (can't update one in place) so we have to
    # use an empty TTFont and then update our GSUB with the newly built lookups
    temp_font = ttLib.TTFont()
    temp_font.setGlyphOrder(colr_font.getGlyphOrder())

    addOpenTypeFeaturesFromString(temp_font, features)

    temp_gsub = temp_font["GSUB"].table
    # sanity check
    assert len(temp_gsub.FeatureList.FeatureRecord) == 1
    assert temp_gsub.FeatureList.FeatureRecord[0].FeatureTag == "ccmp"
    temp_ccmp = temp_gsub.FeatureList.FeatureRecord[0].Feature

    colr_gsub = colr_font["GSUB"].table
    ccmps = [
        r.Feature for r in colr_gsub.FeatureList.FeatureRecord if r.FeatureTag == "ccmp"
    ]
    assert len(ccmps) == 1, f"expected only 1 'ccmp' feature record, found {len(ccmps)}"
    colr_ccmp = ccmps[0]

    colr_lookups = colr_gsub.LookupList.Lookup
    assert (
        len(colr_lookups) == 1
    ), f"expected only 1 lookup in COLRv1's GSUB.LookupList, found {len(colr_lookups)}"
    assert (
        colr_lookups[0].LookupType == 4
    ), f"expected Lookup[0] of type 4 in COLRv1, found {colr_lookups[0].LookupType}"

    colr_lookups.extend(temp_gsub.LookupList.Lookup[1:])
    colr_gsub.LookupList.LookupCount = len(colr_lookups)
    colr_ccmp.LookupListIndex = temp_ccmp.LookupListIndex
    colr_ccmp.LookupCount = len(colr_ccmp.LookupListIndex)

    # get rid of the Unknown Flag private codepoint as no longer needed
    font_data.delete_from_cmap(colr_font, [UNKNOWN_FLAG_PUA])


def _set_no_font_embedding_restrictions(colr_font):
    # The CBDT/CBLC NotoColorEmoji has OS/2.fsType = 0 (i.e. no embedding restrictions)
    # so the COLRv1 variant must also have no such restrictions.
    # https://github.com/notofonts/noto-fonts/issues/2408
    # https://github.com/google/fonts/issues/5729
    colr_font["OS/2"].fsType = 0


def _set_head_version_to_name_version(colr_font):
    # head.fontRevision and the version on name 5 should match
    name_version = colr_font['name'].getName(5, 3, 1, 0x409)
    assert name_version is not None, "No version found in 'name'"
    name_version = name_version.toUnicode()
    match = re.match(r'^Version (\d+[.]\d+);GOOG;', name_version)
    assert match is not None, f"Unable to parse version from '{name_version}'"
    colr_font["head"].fontRevision = float(match.group(1))


def _font(path, check_fn, check_fail_str):
    assert path.is_file(), path
    font = ttLib.TTFont(path)
    if not check_fn(font):
        raise ValueError(path + check_fail_str)
    return font


def main(_):
    cbdt_font = _font(_CBDT_FILE, _is_cbdt, " must be a CBDT font")

    for colr_file in _COLR_FILES:
        colr_font = _font(colr_file, _is_colrv1, " must be a COLRv1 font")

        print(f"Updating {colr_file} from {_CBDT_FILE}")

        _copy_names(colr_font, cbdt_font)

        # CBDT build step: @$(PYTHON) $(PUA_ADDER) "$@" "$@-with-pua"
        map_pua_emoji.add_pua_cmap_to_font(colr_font)

        _add_vs_cmap(colr_font)

        _map_missing_flag_tag_chars_to_empty_glyphs(colr_font)

        add_soft_light_to_flags(colr_font)

        _add_vertical_layout_tables(cbdt_font, colr_font)

        _add_fallback_subs_for_unknown_flags(colr_font)

        _set_no_font_embedding_restrictions(colr_font)

        _set_head_version_to_name_version(colr_font)

        print("Writing", colr_file)
        colr_font.save(colr_file)


if __name__ == "__main__":
    app.run(main)