""" Tests for `telethon.extensions.markdown`. """ from telethon.extensions import markdown from telethon.tl.types import MessageEntityBold, MessageEntityItalic, MessageEntityTextUrl def test_entity_edges(): """ Test that entities at the edges (start and end) don't crash. """ text = 'Hello, world' entities = [MessageEntityBold(0, 5), MessageEntityBold(7, 5)] result = markdown.unparse(text, entities) assert result == '**Hello**, **world**' def test_malformed_entities(): """ Test that malformed entity offsets from bad clients don't crash and produce the expected results. """ text = '🏆Telegram Official Android Challenge is over🏆.' entities = [MessageEntityTextUrl(offset=2, length=43, url='https://example.com')] result = markdown.unparse(text, entities) assert result == "🏆[Telegram Official Android Challenge is over🏆](https://example.com)." def test_trailing_malformed_entities(): """ Similar to `test_malformed_entities`, but for the edge case where the malformed entity offset is right at the end (note the lack of a trailing dot in the text string). """ text = '🏆Telegram Official Android Challenge is over🏆' entities = [MessageEntityTextUrl(offset=2, length=43, url='https://example.com')] result = markdown.unparse(text, entities) assert result == "🏆[Telegram Official Android Challenge is over🏆](https://example.com)" def test_entities_together(): """ Test that an entity followed immediately by a different one behaves well. """ original = '**⚙️**__Settings__' stripped = '⚙️Settings' text, entities = markdown.parse(original) assert text == stripped assert entities == [MessageEntityBold(0, 2), MessageEntityItalic(2, 8)] text = markdown.unparse(text, entities) assert text == original