def test_umarkdown_with_smart(): assert "<p>Hello---</p>\n" == markdown("Hello---") assert "<p>Hello—</p>\n" == markdown("Hello---", smart=True)
def test_umarkdown_with_text(): with pytest.raises(TypeError): markdown() with pytest.raises(TypeError): markdown(text_file="not_exists.md") assert "<h1>Hello World</h1>\n" == markdown(text="# Hello World")
def test_umarkdown_with_no_breaks(): assert "<p>Hello,\nWorld!</p>\n" == markdown("Hello,\nWorld!") assert "<p>Hello, World!</p>\n" == markdown("Hello,\nWorld!", no_breaks=True)
def test_umarkdown_with_unsafe(): assert "<!-- raw HTML omitted -->\n" == markdown("<p>Hello World!</p>") assert "<p>Hello World!</p>\n" == markdown("<p>Hello World!</p>", unsafe=True)
def test_umarkdown_with_hard_breaks(): assert "<p>Hello,\nWorld!</p>\n" == markdown("Hello,\nWorld!") assert "<p>Hello,<br />\nWorld!</p>\n" == markdown("Hello,\nWorld!", hard_breaks=True)
def test_umarkdown_with_source_pos(): assert '<h1 data-sourcepos="1:1-1:13">Hello World</h1>\n' == markdown( "# Hello World", source_pos=True)
def test_umarkdown_with_text_file_and_output_file(): with open("hello.md", "w") as f: f.write("# Hello World") assert markdown(text_file="hello.md", output_file="hello.html") assert "<h1>Hello World</h1>\n" == Path("hello.html").read_text()
def test_umarkdown_with_output_file(): markdown("# Hello World", output_file="hello.html") assert "<h1>Hello World</h1>\n" == Path("hello.html").read_text()
def test_umarkdown_with_text_file(): with open("hello.md", "w") as f: f.write("# Hello World") assert "<h1>Hello World</h1>\n" == markdown(text_file="hello.md")