zulip/tools/setup/emoji/emoji_setup_utils.py

319 lines
15 KiB
Python

# This tool contains all of the rules that we use to decide which of
# the various emoji names in emoji-map.json we should actually use in
# autocomplete and emoji pickers. You can't do all of them, because
# otherwise there will be a ton of duplicates alphabetically next to
# each other, which is confusing and looks bad (e.g. `angry` and
# `angry_face` or `ab` and `ab_button` will always sort next to each
# other, and you really want to just pick one). See docs/emoji.md for
# details on how this system works.
from collections import defaultdict
from itertools import permutations, chain
import ujson
from six.moves import range, zip
from typing import Any, Dict, List, Text
# Emojisets that we currently support.
EMOJISETS = ['apple', 'emojione', 'google', 'twitter']
# the corresponding code point will be set to exactly these names as a
# final pass, overriding any other rules. This is useful for cases
# where the two names are very different, users might reasonably type
# either name and be surprised when they can't find the relevant emoji.
whitelisted_names = [
['date', 'calendar'], ['shirt', 'tshirt'], ['cupid', 'heart_with_arrow'],
['tada', 'party_popper'], ['parking', 'p_button'], ['car', 'automobile'],
['mortar_board', 'graduation_cap'], ['cd', 'optical_disc'], ['tv', 'television'],
['sound', 'speaker_on'], ['mute', 'speaker_off'], ['antenna_bars', 'signal_strength'],
['mag_right', 'right_pointing_magnifying_glass'], ['mag', 'left_pointing_magnifying_glass'],
['loud_sound', 'speaker_loud'], ['rice_scene', 'moon_ceremony'],
['fast_up_button', 'arrow_double_up'], ['fast_down_button', 'arrow_double_down'],
['rewind', 'fast_reverse_button'], ['100', 'hundred_points'], ['muscle', 'flexed_biceps'],
['walking', 'pedestrian'], ['email', 'envelope'], ['dart', 'direct_hit'],
['wc', 'water_closet'], ['zap', 'high_voltage'], ['underage', 'no_one_under_eighteen'],
['vhs', 'videocassette'], ['bangbang', 'double_exclamation_mark'],
['gun', 'pistol'], ['hocho', 'kitchen_knife'], ['8ball', 'billiards'],
['pray', 'folded_hands'], ['cop', 'police_officer'], ['phone', 'telephone'],
['bee', 'honeybee'], ['lips', 'mouth'], ['boat', 'sailboat'], ['feet', 'paw_prints'],
['uk', 'gb'], ['alien_monster', 'space_invader'], ['reverse_button', 'arrow_backward'],
# both github and slack remove play_button, though I think this is better
['play_button', 'arrow_forward'],
# github/slack both get rid of shuffle_tracks_button, which seems wrong
['shuffle_tracks_button', 'twisted_rightwards_arrows'],
['iphone', 'mobile_phone'], # disagrees with github/slack/emojione
# both github and slack remove {growing,beating}_heart, not sure what I think
['heartpulse', 'growing_heart'], ['heartbeat', 'beating_heart'],
# did remove cityscape_at_dusk from (city_sunset, cityscape_at_dusk)
['sunset', 'city_sunrise'],
['punch', 'oncoming_fist'], # doesn't include facepunch
['+1', 'thumbs_up'], # doesn't include thumbsup
['-1', 'thumbs_down'], # doesn't include thumbsdown
# shit, hankey. slack allows poop, shit, hankey. github calls it hankey,
# and autocompletes for poop and shit. emojione calls it poop, and
# autocompletes for pile_of_poo and shit.
['poop', 'pile_of_poo'],
# github/slack remove cooking, but their emoji for this is an uncooked egg
['egg', 'cooking'],
# to match two_{men,women}_holding_hands
['couple', 'man_and_woman_holding_hands'],
# ['ocean', 'water_wave'], wave is so common that we want it to point only to :wave:
]
# We blacklist certain names in cases where the algorithms below would
# choose incorrectly which one to keep. For example, with `football`,
# by default, our algorithm would pick just `football`, but we given
# that :rugby_football: also exists, we want to keep
# :american_football: instead. So we just remove the shorter names here.
blacklisted_names = frozenset([
# would be chosen by words_supersets or superstrings
'football', # american_football
'post_office', # european_post_office (there's also a japanese_post_office)
'castle', # european_castle (there's also a japanese_castle)
'chart', # chart_increasing_with_yen (should rename chart_increasing to chart)
'loop', # double_curly_loop (should rename curly_loop to loop)
'massage', # face_massage
'bulb', # light_bulb
'barber', # barber_pole
'mens', # mens_room
'womens', # womens_room
'knife', # kitchen_knife (hocho also maps here)
'notes', # musical_notes
'beetle', # lady_beetle
'ab', # ab_button (due to keeping a_button, due to the one_lettered() rule)
'headphone', # headphones
'mega', # megaphone
'ski', # skis
'high_heel', # high_heeled_shoe (so that it shows up when searching for shoe)
# less confident about the following
'dolls', # japanese_dolls
'moon', # waxing_gibbous_moon (should rename crescent_moon to moon)
'clapper', # clapper_board
'traffic_light', # horizontal_traffic_light (there's also a vertical_traffic_light)
'lantern',
'red_paper_lantern', # izakaya_lantern (in the future we should make sure
# red_paper_lantern finds this)
# would be chosen by longer
'down_button', # arrow_down_small, I think to match the other arrow_*
# names. Matching what github and slack do.
'running_shoe', # athletic_shoe, both github and slack agree here.
'running', # runner. slack has both, github has running_man and running_woman, but not runner
'o2', # o_button
'star2', # glowing_star
'bright', # high_brightness, to match low_brightness, what github/slack do
'dim_button', # low_brightness, copying github/slack
'stars', # shooting_star. disagrees with github, slack, and emojione, but this seems better
'nail_care', # nail_polish. Also disagrees github/slack/emojione, is nail_polish mostly an
# american thing?
'busstop', # bus_stop
'tophat', # top_hat
'old_woman', # older_woman, following github/slack/emojione on these
'old_man', # older_man
'blue_car', # recreational_vehicle
'litter_in_bin_sign', # put_litter_in_its_place
'moai', # moyai based on github/slack
'fuelpump', # fuel_pump
# names not otherwise excluded by our heuristics
'left_arrow', # arrow_left, to match other arrow_* shortnames
'right_arrow', # arrow_right
'up_arrow', # arrow_up
'down_arrow', # arrow_down
'chequered_flag', # checkered_flag
'e_mail', # e-mail
'non_potable_water', # non-potable_water
'flipper', # dolphin
])
## functions that take in a list of names at a codepoint and return a subset to exclude
def blacklisted(names):
# type: (List[str]) -> List[str]
return [name for name in names if name in blacklisted_names]
# 1 letter names don't currently show up in our autocomplete. Maybe should
# change our autocomplete so that a whitelist of letters do, like j (for joy), x, etc
# github uses a, ab, etc. instead of a_button, slack doesn't have any of the [letter]_buttons
def one_lettered(names):
# type: (List[str]) -> List[str]
if len(names) == 1:
return []
return [name for name in names if len(name) == 1]
# If it is an ideograph (or katakana, but we'll probably deal with that
# differently after 1.5), remove any names that don't have
# ideograph/katakana in them
def ideographless(names):
# type: (List[str]) -> List[str]
has_ideographs = ['ideograph' in name.split('_') or
'katakana' in name.split('_') for name in names]
if not any(has_ideographs):
return []
return [name for name, has_ideograph in zip(names, has_ideographs) if not has_ideograph]
# In the absence of a good reason not to, we prefer :angry: over
# :angry_face:, since it's shorter and communicates the same idea.
#
# This rule is subsumed by the longer rule, but still useful for
# breaking up a hand review of the whitelist/blacklist decisions,
# since these cases are much more clear than the "longer" ones.
def word_superset(names):
# type: (List[str]) -> List[str]
bags_of_words = [frozenset(name.split('_')) for name in names]
bad_names = set()
for i, j in permutations(list(range(len(names))), 2):
if bags_of_words[i] < bags_of_words[j]:
bad_names.add(names[j])
return list(bad_names)
# We prefer :dog: over :dog2: if they both point to the same unicode
# character.
#
# This rule is subsumed by the longer rule, but still useful for
# breaking up a hand review of the whitelist/blacklist decisions,
# since these cases are much more clear than the "longer" ones.
def superstring(names):
# type: (List[str]) -> List[str]
bad_names = set()
for name1, name2 in permutations(names, 2):
if name2[:len(name1)] == name1:
bad_names.add(name2)
return list(bad_names)
# The shorter one is usually a better name.
def longer(names):
# type: (List[str]) -> List[str]
lengths = [len(name) for name in names]
min_length = min(lengths)
return [name for name, length in zip(names, lengths) if length > min_length]
# A lot of emoji that have a color in their name aren't actually the
# right color, which is super confusing. A big part of the reason is
# that "black" and "white" actually mean filled-in and not-filled-in
# to the Unicode committee, which is a poor choice by explains why
# something with "black" in its name might be any solid color. Users
# want the emoji to have reasonable names, though, so we have to
# correct the names with "black" or "white" in them.
#
# Ones found after a few minutes of inspection, and after all the other filters
# have been applied. Probably others remaining.
miscolored_names = frozenset(['eight_pointed_black_star', 'large_blue_diamond',
'small_blue_diamond'])
def google_color_bug(names):
# type: (List[str]) -> List[str]
return [name for name in names if
name[:5] == 'black' or name[:5] == 'white' or name in miscolored_names]
def emoji_names_for_picker(emoji_map):
# type: (Dict[Text, Text]) -> List[str]
codepoint_to_names = defaultdict(list) # type: Dict[Text, List[str]]
for name, codepoint in emoji_map.items():
codepoint_to_names[codepoint].append(str(name))
# blacklisted must come first, followed by {one_lettered, ideographless}
# Each function here returns a list of names to be removed from a list of names
for func in [blacklisted, one_lettered, ideographless, word_superset,
superstring, longer, google_color_bug]:
for codepoint, names in codepoint_to_names.items():
codepoint_to_names[codepoint] = [name for name in names if name not in func(names)]
for names in whitelisted_names:
codepoint = emoji_map[names[0]]
for name in names:
assert (emoji_map[name] == codepoint)
codepoint_to_names[codepoint] = names
return sorted(list(chain.from_iterable(codepoint_to_names.values())))
# Returns a dict from categories to list of codepoints. The list of
# codepoints are sorted according to the `sort_order` as defined in
# `emoji_data`.
def generate_emoji_catalog(emoji_data):
# type: (List[Dict[Text, Any]]) -> Dict[str, List[str]]
sort_order = {} # type: Dict[str, int]
emoji_catalog = {} # type: Dict[str, List[str]]
for emoji in emoji_data:
if not emoji_is_universal(emoji):
continue
category = str(emoji["category"])
codepoint = str(emoji["unified"]).lower()
sort_order[codepoint] = emoji["sort_order"]
if category in emoji_catalog:
emoji_catalog[category].append(codepoint)
else:
emoji_catalog[category] = [codepoint, ]
for category in emoji_catalog:
emoji_catalog[category].sort(key=lambda codepoint: sort_order[codepoint])
return emoji_catalog
# Use only those names for which images are present in all
# the emoji sets so that we can switch emoji sets seemlessly.
def emoji_is_universal(emoji_dict):
# type: (Dict[Text, Any]) -> bool
for emoji_set in EMOJISETS:
if not emoji_dict['has_img_' + emoji_set]:
return False
return True
def generate_codepoint_to_name_map(names, unified_reactions_data):
# type: (List[str], Dict[Text, Text]) -> Dict[str, str]
# TODO: Decide canonical names. For now, using the names
# generated for emoji picker. In case of multiple names
# for the same emoji, lexicographically greater name is
# used, for example, `thumbs_up` is used and not `+1`.
codepoint_to_name = {} # type: Dict[str, str]
for name in names:
codepoint_to_name[str(unified_reactions_data[name])] = str(name)
return codepoint_to_name
def emoji_can_be_included(emoji_dict, unified_reactions_codepoints):
# type: (Dict[Text, Any], List[Text]) -> bool
# This function returns True if an emoji in new(not included in old emoji dataset) and is
# safe to be included. Currently emojis which are represented by a sequence of codepoints
# or emojis with ZWJ are not to be included until we implement a mechanism for dealing with
# their unicode versions.
# `:fried_egg:` emoji is banned for now, due to a name collision with `:egg:` emoji in
# `unified_reactions.json` dataset, until we completely switch to iamcal dataset.
if emoji_dict["short_name"] == "fried_egg":
return False
codepoint = emoji_dict["unified"].lower()
if '-' not in codepoint and emoji_dict["category"] != "Skin Tones" and \
emoji_is_universal(emoji_dict) and codepoint not in unified_reactions_codepoints:
return True
return False
def get_new_emoji_dicts(unified_reactions_data, emoji_data):
# type: (Dict[Text, Text], List[Dict[Text, Any]]) -> List[Dict[Text, Any]]
unified_reactions_codepoints = [unified_reactions_data[name] for name in unified_reactions_data]
new_emoji_dicts = []
for emoji_dict in emoji_data:
if emoji_can_be_included(emoji_dict, unified_reactions_codepoints):
new_emoji_dicts.append(emoji_dict)
return new_emoji_dicts
def get_extended_names_list(names, new_emoji_dicts):
# type: (List[Text], List[Dict[Text, Any]]) -> List[Text]
extended_names_list = names[:]
for emoji_dict in new_emoji_dicts:
extended_names_list.append(emoji_dict["short_name"])
return extended_names_list
def get_extended_name_to_codepoint(name_to_codepoint, new_emoji_dicts):
# type: (Dict[Text, Text], List[Dict[Text, Any]]) -> Dict[Text, Text]
extended_name_to_codepoint = name_to_codepoint.copy()
for emoji_dict in new_emoji_dicts:
emoji_name = emoji_dict["short_name"]
codepoint = emoji_dict["unified"].lower()
extended_name_to_codepoint[emoji_name] = codepoint
return extended_name_to_codepoint
def get_extended_codepoint_to_name(codepoint_to_name, new_emoji_dicts):
# type: (Dict[Text, Text], List[Dict[Text, Any]]) -> Dict[Text, Text]
extended_codepoint_to_name = codepoint_to_name.copy()
for emoji_dict in new_emoji_dicts:
emoji_name = emoji_dict["short_name"]
codepoint = emoji_dict["unified"].lower()
extended_codepoint_to_name[codepoint] = emoji_name
return extended_codepoint_to_name