You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
253 lines
12 KiB
253 lines
12 KiB
from fontTools import unicodedata
|
|
|
|
import pytest
|
|
|
|
|
|
def test_script():
|
|
assert unicodedata.script("a") == "Latn"
|
|
assert unicodedata.script(chr(0)) == "Zyyy"
|
|
assert unicodedata.script(chr(0x0378)) == "Zzzz"
|
|
assert unicodedata.script(chr(0x10FFFF)) == "Zzzz"
|
|
|
|
# these were randomly sampled, one character per script
|
|
assert unicodedata.script(chr(0x1E918)) == 'Adlm'
|
|
assert unicodedata.script(chr(0x1170D)) == 'Ahom'
|
|
assert unicodedata.script(chr(0x145A0)) == 'Hluw'
|
|
assert unicodedata.script(chr(0x0607)) == 'Arab'
|
|
assert unicodedata.script(chr(0x056C)) == 'Armn'
|
|
assert unicodedata.script(chr(0x10B27)) == 'Avst'
|
|
assert unicodedata.script(chr(0x1B41)) == 'Bali'
|
|
assert unicodedata.script(chr(0x168AD)) == 'Bamu'
|
|
assert unicodedata.script(chr(0x16ADD)) == 'Bass'
|
|
assert unicodedata.script(chr(0x1BE5)) == 'Batk'
|
|
assert unicodedata.script(chr(0x09F3)) == 'Beng'
|
|
assert unicodedata.script(chr(0x11C5B)) == 'Bhks'
|
|
assert unicodedata.script(chr(0x3126)) == 'Bopo'
|
|
assert unicodedata.script(chr(0x1103B)) == 'Brah'
|
|
assert unicodedata.script(chr(0x2849)) == 'Brai'
|
|
assert unicodedata.script(chr(0x1A0A)) == 'Bugi'
|
|
assert unicodedata.script(chr(0x174E)) == 'Buhd'
|
|
assert unicodedata.script(chr(0x18EE)) == 'Cans'
|
|
assert unicodedata.script(chr(0x102B7)) == 'Cari'
|
|
assert unicodedata.script(chr(0x1053D)) == 'Aghb'
|
|
assert unicodedata.script(chr(0x11123)) == 'Cakm'
|
|
assert unicodedata.script(chr(0xAA1F)) == 'Cham'
|
|
assert unicodedata.script(chr(0xAB95)) == 'Cher'
|
|
assert unicodedata.script(chr(0x1F0C7)) == 'Zyyy'
|
|
assert unicodedata.script(chr(0x2C85)) == 'Copt'
|
|
assert unicodedata.script(chr(0x12014)) == 'Xsux'
|
|
assert unicodedata.script(chr(0x1082E)) == 'Cprt'
|
|
assert unicodedata.script(chr(0xA686)) == 'Cyrl'
|
|
assert unicodedata.script(chr(0x10417)) == 'Dsrt'
|
|
assert unicodedata.script(chr(0x093E)) == 'Deva'
|
|
assert unicodedata.script(chr(0x1BC4B)) == 'Dupl'
|
|
assert unicodedata.script(chr(0x1310C)) == 'Egyp'
|
|
assert unicodedata.script(chr(0x1051C)) == 'Elba'
|
|
assert unicodedata.script(chr(0x2DA6)) == 'Ethi'
|
|
assert unicodedata.script(chr(0x10AD)) == 'Geor'
|
|
assert unicodedata.script(chr(0x2C52)) == 'Glag'
|
|
assert unicodedata.script(chr(0x10343)) == 'Goth'
|
|
assert unicodedata.script(chr(0x11371)) == 'Gran'
|
|
assert unicodedata.script(chr(0x03D0)) == 'Grek'
|
|
assert unicodedata.script(chr(0x0AAA)) == 'Gujr'
|
|
assert unicodedata.script(chr(0x0A4C)) == 'Guru'
|
|
assert unicodedata.script(chr(0x23C9F)) == 'Hani'
|
|
assert unicodedata.script(chr(0xC259)) == 'Hang'
|
|
assert unicodedata.script(chr(0x1722)) == 'Hano'
|
|
assert unicodedata.script(chr(0x108F5)) == 'Hatr'
|
|
assert unicodedata.script(chr(0x05C2)) == 'Hebr'
|
|
assert unicodedata.script(chr(0x1B072)) == 'Hira'
|
|
assert unicodedata.script(chr(0x10847)) == 'Armi'
|
|
assert unicodedata.script(chr(0x033A)) == 'Zinh'
|
|
assert unicodedata.script(chr(0x10B66)) == 'Phli'
|
|
assert unicodedata.script(chr(0x10B4B)) == 'Prti'
|
|
assert unicodedata.script(chr(0xA98A)) == 'Java'
|
|
assert unicodedata.script(chr(0x110B2)) == 'Kthi'
|
|
assert unicodedata.script(chr(0x0CC6)) == 'Knda'
|
|
assert unicodedata.script(chr(0x3337)) == 'Kana'
|
|
assert unicodedata.script(chr(0xA915)) == 'Kali'
|
|
assert unicodedata.script(chr(0x10A2E)) == 'Khar'
|
|
assert unicodedata.script(chr(0x17AA)) == 'Khmr'
|
|
assert unicodedata.script(chr(0x11225)) == 'Khoj'
|
|
assert unicodedata.script(chr(0x112B6)) == 'Sind'
|
|
assert unicodedata.script(chr(0x0ED7)) == 'Laoo'
|
|
assert unicodedata.script(chr(0xAB3C)) == 'Latn'
|
|
assert unicodedata.script(chr(0x1C48)) == 'Lepc'
|
|
assert unicodedata.script(chr(0x1923)) == 'Limb'
|
|
assert unicodedata.script(chr(0x1071D)) == 'Lina'
|
|
assert unicodedata.script(chr(0x100EC)) == 'Linb'
|
|
assert unicodedata.script(chr(0xA4E9)) == 'Lisu'
|
|
assert unicodedata.script(chr(0x10284)) == 'Lyci'
|
|
assert unicodedata.script(chr(0x10926)) == 'Lydi'
|
|
assert unicodedata.script(chr(0x11161)) == 'Mahj'
|
|
assert unicodedata.script(chr(0x0D56)) == 'Mlym'
|
|
assert unicodedata.script(chr(0x0856)) == 'Mand'
|
|
assert unicodedata.script(chr(0x10AF0)) == 'Mani'
|
|
assert unicodedata.script(chr(0x11CB0)) == 'Marc'
|
|
assert unicodedata.script(chr(0x11D28)) == 'Gonm'
|
|
assert unicodedata.script(chr(0xABDD)) == 'Mtei'
|
|
assert unicodedata.script(chr(0x1E897)) == 'Mend'
|
|
assert unicodedata.script(chr(0x109B0)) == 'Merc'
|
|
assert unicodedata.script(chr(0x10993)) == 'Mero'
|
|
assert unicodedata.script(chr(0x16F5D)) == 'Plrd'
|
|
assert unicodedata.script(chr(0x1160B)) == 'Modi'
|
|
assert unicodedata.script(chr(0x18A8)) == 'Mong'
|
|
assert unicodedata.script(chr(0x16A48)) == 'Mroo'
|
|
assert unicodedata.script(chr(0x1128C)) == 'Mult'
|
|
assert unicodedata.script(chr(0x105B)) == 'Mymr'
|
|
assert unicodedata.script(chr(0x108AF)) == 'Nbat'
|
|
assert unicodedata.script(chr(0x19B3)) == 'Talu'
|
|
assert unicodedata.script(chr(0x1143D)) == 'Newa'
|
|
assert unicodedata.script(chr(0x07F4)) == 'Nkoo'
|
|
assert unicodedata.script(chr(0x1B192)) == 'Nshu'
|
|
assert unicodedata.script(chr(0x169C)) == 'Ogam'
|
|
assert unicodedata.script(chr(0x1C56)) == 'Olck'
|
|
assert unicodedata.script(chr(0x10CE9)) == 'Hung'
|
|
assert unicodedata.script(chr(0x10316)) == 'Ital'
|
|
assert unicodedata.script(chr(0x10A93)) == 'Narb'
|
|
assert unicodedata.script(chr(0x1035A)) == 'Perm'
|
|
assert unicodedata.script(chr(0x103D5)) == 'Xpeo'
|
|
assert unicodedata.script(chr(0x10A65)) == 'Sarb'
|
|
assert unicodedata.script(chr(0x10C09)) == 'Orkh'
|
|
assert unicodedata.script(chr(0x0B60)) == 'Orya'
|
|
assert unicodedata.script(chr(0x104CF)) == 'Osge'
|
|
assert unicodedata.script(chr(0x104A8)) == 'Osma'
|
|
assert unicodedata.script(chr(0x16B12)) == 'Hmng'
|
|
assert unicodedata.script(chr(0x10879)) == 'Palm'
|
|
assert unicodedata.script(chr(0x11AF1)) == 'Pauc'
|
|
assert unicodedata.script(chr(0xA869)) == 'Phag'
|
|
assert unicodedata.script(chr(0x10909)) == 'Phnx'
|
|
assert unicodedata.script(chr(0x10B81)) == 'Phlp'
|
|
assert unicodedata.script(chr(0xA941)) == 'Rjng'
|
|
assert unicodedata.script(chr(0x16C3)) == 'Runr'
|
|
assert unicodedata.script(chr(0x0814)) == 'Samr'
|
|
assert unicodedata.script(chr(0xA88C)) == 'Saur'
|
|
assert unicodedata.script(chr(0x111C8)) == 'Shrd'
|
|
assert unicodedata.script(chr(0x1045F)) == 'Shaw'
|
|
assert unicodedata.script(chr(0x115AD)) == 'Sidd'
|
|
assert unicodedata.script(chr(0x1D8C0)) == 'Sgnw'
|
|
assert unicodedata.script(chr(0x0DB9)) == 'Sinh'
|
|
assert unicodedata.script(chr(0x110F9)) == 'Sora'
|
|
assert unicodedata.script(chr(0x11A60)) == 'Soyo'
|
|
assert unicodedata.script(chr(0x1B94)) == 'Sund'
|
|
assert unicodedata.script(chr(0xA81F)) == 'Sylo'
|
|
assert unicodedata.script(chr(0x0740)) == 'Syrc'
|
|
assert unicodedata.script(chr(0x1714)) == 'Tglg'
|
|
assert unicodedata.script(chr(0x1761)) == 'Tagb'
|
|
assert unicodedata.script(chr(0x1965)) == 'Tale'
|
|
assert unicodedata.script(chr(0x1A32)) == 'Lana'
|
|
assert unicodedata.script(chr(0xAA86)) == 'Tavt'
|
|
assert unicodedata.script(chr(0x116A5)) == 'Takr'
|
|
assert unicodedata.script(chr(0x0B8E)) == 'Taml'
|
|
assert unicodedata.script(chr(0x1754D)) == 'Tang'
|
|
assert unicodedata.script(chr(0x0C40)) == 'Telu'
|
|
assert unicodedata.script(chr(0x07A4)) == 'Thaa'
|
|
assert unicodedata.script(chr(0x0E42)) == 'Thai'
|
|
assert unicodedata.script(chr(0x0F09)) == 'Tibt'
|
|
assert unicodedata.script(chr(0x2D3A)) == 'Tfng'
|
|
assert unicodedata.script(chr(0x114B0)) == 'Tirh'
|
|
assert unicodedata.script(chr(0x1038B)) == 'Ugar'
|
|
assert unicodedata.script(chr(0xA585)) == 'Vaii'
|
|
assert unicodedata.script(chr(0x118CF)) == 'Wara'
|
|
assert unicodedata.script(chr(0xA066)) == 'Yiii'
|
|
assert unicodedata.script(chr(0x11A31)) == 'Zanb'
|
|
|
|
|
|
def test_script_extension():
|
|
assert unicodedata.script_extension("a") == {"Latn"}
|
|
assert unicodedata.script_extension(chr(0)) == {"Zyyy"}
|
|
assert unicodedata.script_extension(chr(0x0378)) == {"Zzzz"}
|
|
assert unicodedata.script_extension(chr(0x10FFFF)) == {"Zzzz"}
|
|
|
|
assert unicodedata.script_extension("\u0660") == {'Arab', 'Thaa', 'Yezi'}
|
|
assert unicodedata.script_extension("\u0964") == {
|
|
'Beng', 'Deva', 'Dogr', 'Gong', 'Gonm', 'Gran', 'Gujr', 'Guru', 'Knda',
|
|
'Mahj', 'Mlym', 'Nand', 'Orya', 'Sind', 'Sinh', 'Sylo', 'Takr', 'Taml',
|
|
'Telu', 'Tirh'}
|
|
|
|
|
|
def test_script_name():
|
|
assert unicodedata.script_name("Latn") == "Latin"
|
|
assert unicodedata.script_name("Zyyy") == "Common"
|
|
assert unicodedata.script_name("Zzzz") == "Unknown"
|
|
# underscores in long names are replaced by spaces
|
|
assert unicodedata.script_name("Egyp") == "Egyptian Hieroglyphs"
|
|
|
|
with pytest.raises(KeyError):
|
|
unicodedata.script_name("QQQQ")
|
|
assert unicodedata.script_name("QQQQ", default="Unknown")
|
|
|
|
|
|
def test_script_code():
|
|
assert unicodedata.script_code("Latin") == "Latn"
|
|
assert unicodedata.script_code("Common") == "Zyyy"
|
|
assert unicodedata.script_code("Unknown") == "Zzzz"
|
|
# case, whitespace, underscores and hyphens are ignored
|
|
assert unicodedata.script_code("Egyptian Hieroglyphs") == "Egyp"
|
|
assert unicodedata.script_code("Egyptian_Hieroglyphs") == "Egyp"
|
|
assert unicodedata.script_code("egyptianhieroglyphs") == "Egyp"
|
|
assert unicodedata.script_code("Egyptian-Hieroglyphs") == "Egyp"
|
|
|
|
with pytest.raises(KeyError):
|
|
unicodedata.script_code("Does not exist")
|
|
assert unicodedata.script_code("Does not exist", default="Zzzz") == "Zzzz"
|
|
|
|
|
|
def test_block():
|
|
assert unicodedata.block("\x00") == "Basic Latin"
|
|
assert unicodedata.block("\x7F") == "Basic Latin"
|
|
assert unicodedata.block("\x80") == "Latin-1 Supplement"
|
|
assert unicodedata.block("\u1c90") == "Georgian Extended"
|
|
assert unicodedata.block("\u0870") == "No_Block"
|
|
|
|
|
|
def test_ot_tags_from_script():
|
|
# simple
|
|
assert unicodedata.ot_tags_from_script("Latn") == ["latn"]
|
|
# script mapped to multiple new and old script tags
|
|
assert unicodedata.ot_tags_from_script("Deva") == ["dev2", "deva"]
|
|
# exceptions
|
|
assert unicodedata.ot_tags_from_script("Hira") == ["kana"]
|
|
# special script codes map to DFLT
|
|
assert unicodedata.ot_tags_from_script("Zinh") == ["DFLT"]
|
|
assert unicodedata.ot_tags_from_script("Zyyy") == ["DFLT"]
|
|
assert unicodedata.ot_tags_from_script("Zzzz") == ["DFLT"]
|
|
# this is invalid or unknown
|
|
assert unicodedata.ot_tags_from_script("Aaaa") == ["DFLT"]
|
|
|
|
|
|
def test_ot_tag_to_script():
|
|
assert unicodedata.ot_tag_to_script("latn") == "Latn"
|
|
assert unicodedata.ot_tag_to_script("kana") == "Kana"
|
|
assert unicodedata.ot_tag_to_script("DFLT") == None
|
|
assert unicodedata.ot_tag_to_script("aaaa") == None
|
|
assert unicodedata.ot_tag_to_script("beng") == "Beng"
|
|
assert unicodedata.ot_tag_to_script("bng2") == "Beng"
|
|
assert unicodedata.ot_tag_to_script("dev2") == "Deva"
|
|
assert unicodedata.ot_tag_to_script("gjr2") == "Gujr"
|
|
assert unicodedata.ot_tag_to_script("yi ") == "Yiii"
|
|
assert unicodedata.ot_tag_to_script("nko ") == "Nkoo"
|
|
assert unicodedata.ot_tag_to_script("vai ") == "Vaii"
|
|
assert unicodedata.ot_tag_to_script("lao ") == "Laoo"
|
|
assert unicodedata.ot_tag_to_script("yi") == "Yiii"
|
|
|
|
for invalid_value in ("", " ", "z zz", "zzzzz"):
|
|
with pytest.raises(ValueError, match="invalid OpenType tag"):
|
|
unicodedata.ot_tag_to_script(invalid_value)
|
|
|
|
|
|
def test_script_horizontal_direction():
|
|
assert unicodedata.script_horizontal_direction("Latn") == "LTR"
|
|
assert unicodedata.script_horizontal_direction("Arab") == "RTL"
|
|
assert unicodedata.script_horizontal_direction("Thaa") == "RTL"
|
|
|
|
with pytest.raises(KeyError):
|
|
unicodedata.script_horizontal_direction("Azzz")
|
|
assert unicodedata.script_horizontal_direction("Azzz",
|
|
default="LTR") == "LTR"
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import sys
|
|
sys.exit(pytest.main(sys.argv))
|