Files

40 lines
1.2 KiB
Python

# Test the unicode support! 👋
ᚴ=2
assert ᚴ*8 == 16
ᚴ="👋"
c = ᚴ*3
assert c == '👋👋👋'
import unicodedata
assert unicodedata.category('a') == 'Ll'
assert unicodedata.category('A') == 'Lu'
assert unicodedata.name('a') == 'LATIN SMALL LETTER A'
assert unicodedata.lookup('LATIN SMALL LETTER A') == 'a'
assert unicodedata.bidirectional('a') == 'L'
assert unicodedata.normalize('NFC', 'bla') == 'bla'
# testing unicodedata.ucd_3_2_0 for idna
assert "abcСĤ".encode("idna") == b'xn--abc-7sa390b'
# TODO: fix: assert "abc䄣IJ".encode("idna") == b'xn--abcij-zb5f'
# from CPython tests
assert "python.org".encode("idna") == b"python.org"
assert "python.org.".encode("idna") == b"python.org."
assert "pyth\xf6n.org".encode("idna") == b"xn--pythn-mua.org"
assert "pyth\xf6n.org.".encode("idna") == b"xn--pythn-mua.org."
assert b"python.org".decode("idna") == "python.org"
assert b"python.org.".decode("idna") == "python.org."
assert b"xn--pythn-mua.org".decode("idna") == "pyth\xf6n.org"
assert b"xn--pythn-mua.org.".decode("idna") == "pyth\xf6n.org."
# TODO: add east_asian_width and mirrored
# assert unicodedata.ucd_3_2_0.east_asian_width('\u231a') == 'N'
# assert not unicodedata.ucd_3_2_0.mirrored("\u0f3a")