# coding: utf-8 """ comment testing is all about roundtrips these can be done in the "old" way by creating a file.data and file.roundtrip but there is little flexibility in doing that but some things are not easily tested, eog. how a roundtrip changes """ import pytest # type: ignore # NOQA import sys from roundtrip import round_trip, dedent, round_trip_load, round_trip_dump class TestComments: def test_no_end_of_file_eol(self) -> None: """not excluding comments caused some problems if at the end of the file without a newline. First error, then included \0 """ x = """\ - europe: 10 # abc""" round_trip(x, extra='\n') with pytest.raises(AssertionError): round_trip(x, extra='a\n') def test_no_comments(self) -> None: round_trip(""" - europe: 10 - usa: - ohio: 2 - california: 9 """) def test_round_trip_ordering(self) -> None: round_trip(""" a: 1 b: 2 c: 3 b1: 2 b2: 2 d: 4 e: 5 f: 6 """) def test_complex(self) -> None: round_trip(""" - europe: 10 # top - usa: - ohio: 2 - california: 9 # o """) def test_dropped(self) -> None: s = """\ # comment scalar ... """ round_trip(s, 'scalar\n...\n') def test_main_mapping_begin_end(self) -> None: round_trip(""" # C start a # C start b abc: 1 ghi: 2 klm: 3 # C end a # C end b """) def test_reindent(self) -> None: x = """\ a: b: # comment 1 c: 1 # comment 2 """ d = round_trip_load(x) y = round_trip_dump(d, indent=4) assert y == dedent("""\ a: b: # comment 1 c: 1 # comment 2 """) def test_main_mapping_begin_end_items_post(self) -> None: round_trip(""" # C start a # C start b abc: 1 # abc comment ghi: 2 klm: 3 # klm comment # C end a # C end b """) def test_main_sequence_begin_end(self) -> None: round_trip(""" # C start a # C start b - abc - ghi - klm # C end a # C end b """) def test_main_sequence_begin_end_items_post(self) -> None: round_trip(""" # C start a # C start b - abc # abc comment - ghi - klm # klm comment # C end a # C end b """) def test_main_mapping_begin_end_complex(self) -> None: round_trip(""" # C start a # C start b abc: 1 ghi: 2 klm: 3a: alpha 3b: beta # it is all greek to me # C end a # C end b """) def test_09(self) -> None: # 2.9 from the examples in the spec s = """\ hr: # 1998 hr ranking - Mark McGwire - Sammy Sosa rbi: # 1998 rbi ranking - Sammy Sosa - Ken Griffey """ round_trip(s, indent=4, block_seq_indent=2) def test_09a(self) -> None: round_trip(""" hr: # 1998 hr ranking - Mark McGwire - Sammy Sosa rbi: # 1998 rbi ranking - Sammy Sosa - Ken Griffey """) def test_simple_map_middle_comment(self) -> None: round_trip(""" abc: 1 # C 3a # C 3b ghi: 2 """) def test_map_in_map_0(self) -> None: round_trip(""" map1: # comment 1 # comment 2 map2: key1: val1 """) def test_map_in_map_1(self) -> None: # comment is moved from value to key round_trip(""" map1: # comment 1 map2: key1: val1 """) def test_application_arguments(self) -> None: # application configur round_trip(""" args: username: anthon passwd: secret fullname: Anthon van der Neut tmux: session-name: test loop: wait: 10 """) def test_substitute(self) -> None: x = """ args: username: anthon # name passwd: secret # password fullname: Anthon van der Neut tmux: session-name: test loop: wait: 10 """ data = round_trip_load(x) data['args']['passwd'] = 'deleted password' # note the requirement to add spaces for alignment of comment x = x.replace(': secret ', ': deleted password') assert round_trip_dump(data) == dedent(x) def test_set_comment(self) -> None: round_trip(""" !!set # the beginning ? a # next one is B (lowercase) ? b # You see? Promised you. ? c # this is the end """) def test_omap_comment_roundtrip(self) -> None: round_trip(""" !!omap - a: 1 - b: 2 # two - c: 3 # three - d: 4 """) def test_omap_comment_roundtrip_pre_comment(self) -> None: round_trip(""" !!omap - a: 1 - b: 2 # two - c: 3 # three # last one - d: 4 """) def test_non_ascii(self) -> None: round_trip(""" verbosity: 1 # 0 is minimal output, -1 none base_url: http://gopher.net special_indices: [1, 5, 8] also_special: - a - 19 - 32 asia and europe: &asia_europe Turkey: Ankara Russia: Moscow countries: Asia: <<: *asia_europe Japan: Tokyo # 東京 Europe: <<: *asia_europe Spain: Madrid Italy: Rome """) def test_dump_utf8(self) -> None: import ruamel.yaml # NOQA x = dedent("""\ ab: - x # comment - y # more comment """) data = round_trip_load(x) for utf in [True, False]: y = round_trip_dump( data, default_flow_style=False, allow_unicode=utf ) assert y == x def test_dump_unicode_utf8(self) -> None: import ruamel.yaml # NOQA x = dedent("""\ ab: - x # comment - y # more comment """) data = round_trip_load(x) for utf in [True, False]: y = round_trip_dump( data, default_flow_style=False, allow_unicode=utf ) assert y == x def test_mlget_00(self) -> None: x = """\ a: - b: c: 42 - d: f: 196 e: g: 3.14 """ d = round_trip_load(x) assert d.mlget(['a', 1, 'd', 'f'], list_ok=True) == 196 # with pytest.raises(AssertionError): # d.mlget(['a', 1, 'd', 'f']) == 196 class TestInsertPopList: """list insertion is more complex than dict insertion, as you need to move the values to subsequent keys on insert""" @property def ins(self) -> str: return """\ ab: - a # a - b # b - c - d # d de: - 1 - 2 """ def test_insert_0(self) -> None: d = round_trip_load(self.ins) d['ab'].insert(0, 'xyz') y = round_trip_dump(d, indent=2) assert y == dedent("""\ ab: - xyz - a # a - b # b - c - d # d de: - 1 - 2 """) def test_insert_1(self) -> None: d = round_trip_load(self.ins) d['ab'].insert(4, 'xyz') y = round_trip_dump(d, indent=2) assert y == dedent("""\ ab: - a # a - b # b - c - d # d - xyz de: - 1 - 2 """) def test_insert_2(self) -> None: d = round_trip_load(self.ins) d['ab'].insert(1, 'xyz') y = round_trip_dump(d, indent=2) assert y == dedent("""\ ab: - a # a - xyz - b # b - c - d # d de: - 1 - 2 """) def test_pop_0(self) -> None: d = round_trip_load(self.ins) d['ab'].pop(0) y = round_trip_dump(d, indent=2) print(y) assert y == dedent("""\ ab: - b # b - c - d # d de: - 1 - 2 """) def test_pop_1(self) -> None: d = round_trip_load(self.ins) d['ab'].pop(1) y = round_trip_dump(d, indent=2) print(y) assert y == dedent("""\ ab: - a # a - c - d # d de: - 1 - 2 """) def test_pop_2(self) -> None: d = round_trip_load(self.ins) d['ab'].pop(2) y = round_trip_dump(d, indent=2) print(y) assert y == dedent("""\ ab: - a # a - b # b - d # d de: - 1 - 2 """) def test_pop_3(self) -> None: d = round_trip_load(self.ins) d['ab'].pop(3) y = round_trip_dump(d, indent=2) print(y) assert y == dedent("""\ ab: - a # a - b # b - c de: - 1 - 2 """) # inspired by demux' question on stackoverflow # http://stackoverflow.com/a/36970608/1307905 class TestInsertInMapping: @property def ins(self) -> str: return """\ first_name: Art occupation: Architect # This is an occupation comment about: Art Vandelay is a fictional character that George invents... """ def test_insert_at_pos_1(self) -> None: d = round_trip_load(self.ins) d.insert(1, 'last name', 'Vandelay', comment='new key') y = round_trip_dump(d) print(y) assert y == dedent("""\ first_name: Art last name: Vandelay # new key occupation: Architect # This is an occupation comment about: Art Vandelay is a fictional character that George invents... """) def test_insert_at_pos_0(self) -> None: d = round_trip_load(self.ins) d.insert(0, 'last name', 'Vandelay', comment='new key') y = round_trip_dump(d) print(y) assert y == dedent("""\ last name: Vandelay # new key first_name: Art occupation: Architect # This is an occupation comment about: Art Vandelay is a fictional character that George invents... """) def test_insert_at_pos_3(self) -> None: # much more simple if done with appending. d = round_trip_load(self.ins) d.insert(3, 'last name', 'Vandelay', comment='new key') y = round_trip_dump(d) print(y) assert y == dedent("""\ first_name: Art occupation: Architect # This is an occupation comment about: Art Vandelay is a fictional character that George invents... last name: Vandelay # new key """) class TestCommentedMapMerge: def test_in_operator(self) -> None: data = round_trip_load(""" x: &base a: 1 b: 2 c: 3 y: <<: *base k: 4 l: 5 """) assert data['x']['a'] == 1 assert 'a' in data['x'] assert data['y']['a'] == 1 assert 'a' in data['y'] def test_issue_60(self) -> None: data = round_trip_load(""" x: &base a: 1 y: <<: *base """) assert data['x']['a'] == 1 assert data['y']['a'] == 1 assert str(data['y']) == """{'a': 1}""" def test_issue_60_1(self) -> None: data = round_trip_load(""" x: &base a: 1 y: <<: *base b: 2 """) assert data['x']['a'] == 1 assert data['y']['a'] == 1 assert str(data['y']) == """{'b': 2, 'a': 1}""" class TestEmptyLines: # prompted by issue 46 from Alex Harvey def test_issue_46(self) -> None: yaml_str = dedent("""\ --- # Please add key/value pairs in alphabetical order aws_s3_bucket: 'mys3bucket' jenkins_ad_credentials: bind_name: 'CN=svc-AAA-BBB-T,OU=Example,DC=COM,DC=EXAMPLE,DC=Local' bind_pass: 'xxxxyyyy{' """) d = round_trip_load(yaml_str, preserve_quotes=True) y = round_trip_dump(d, explicit_start=True) assert yaml_str == y def test_multispace_map(self) -> None: round_trip(""" a: 1x b: 2x c: 3x d: 4x """) @pytest.mark.xfail(strict=True) # type: ignore def test_multispace_map_initial(self) -> None: round_trip(""" a: 1x b: 2x c: 3x d: 4x """) def test_embedded_map(self) -> None: round_trip(""" - a: 1y b: 2y c: 3y """) def test_toplevel_seq(self) -> None: round_trip("""\ - 1 - 2 - 3 """) def test_embedded_seq(self) -> None: round_trip(""" a: b: - 1 - 2 - 3 """) def test_line_with_only_spaces(self) -> None: # issue 54 yaml_str = "---\n\na: 'x'\n \nb: y\n" d = round_trip_load(yaml_str, preserve_quotes=True) y = round_trip_dump(d, explicit_start=True) stripped = "" for line in yaml_str.splitlines(): stripped += line.rstrip() + '\n' print(line + '$') assert stripped == y def test_some_eol_spaces(self) -> None: # spaces after tokens and on empty lines yaml_str = '--- \n \na: "x" \n \nb: y \n' d = round_trip_load(yaml_str, preserve_quotes=True) y = round_trip_dump(d, explicit_start=True) stripped = "" for line in yaml_str.splitlines(): stripped += line.rstrip() + '\n' print(line + '$') assert stripped == y def test_issue_54_not_ok(self) -> None: yaml_str = dedent("""\ toplevel: # some comment sublevel: 300 """) d = round_trip_load(yaml_str) print(d.ca) y = round_trip_dump(d, indent=4) assert isinstance(y, str) print(y.replace('\n', '$\n')) assert yaml_str == y def test_issue_54_ok(self) -> None: yaml_str = dedent("""\ toplevel: # some comment sublevel: 300 """) d = round_trip_load(yaml_str) y = round_trip_dump(d, indent=4) assert yaml_str == y def test_issue_93(self) -> None: round_trip("""\ a: b: - c1: cat # a1 # my comment on catfish - c2: catfish # a2 """) def test_issue_93_00(self) -> None: round_trip("""\ a: - - c1: cat # a1 # my comment on catfish - c2: catfish # a2 """) def test_issue_93_01(self) -> None: round_trip("""\ - - c1: cat # a1 # my comment on catfish - c2: catfish # a2 """) def test_issue_93_02(self) -> None: # never failed as there is no indent round_trip("""\ - c1: cat # my comment on catfish - c2: catfish """) def test_issue_96(self) -> None: # inserted extra line on trailing spaces round_trip("""\ a: b: c: c_val d: e: g: g_val """) class TestUnicodeComments: @pytest.mark.skipif(sys.version_info < (2, 7), reason='wide unicode') # type: ignore def test_issue_55(self) -> None: # reported by Haraguroicha Hsu round_trip("""\ name: TEST description: test using author: Harguroicha sql: command: |- select name from testtbl where no = :no ci-test: - :no: 04043709 # 小花 - :no: 05161690 # 茶 - :no: 05293147 # 〇𤋥川 - :no: 05338777 # 〇〇啓 - :no: 05273867 # 〇 - :no: 05205786 # 〇𤦌 """) class TestEmptyValueBeforeComments: def test_issue_25a(self) -> None: round_trip("""\ - a: b c: d d: # foo - e: f """) def test_issue_25a1(self) -> None: round_trip("""\ - a: b c: d d: # foo e: f """) def test_issue_25b(self) -> None: round_trip("""\ var1: #empty var2: something #notempty """) def test_issue_25c(self) -> None: round_trip("""\ params: a: 1 # comment a b: # comment b c: 3 # comment c """) def test_issue_25c1(self) -> None: round_trip("""\ params: a: 1 # comment a b: # comment b # extra c: 3 # comment c """) def test_issue_25_00(self) -> None: round_trip("""\ params: a: 1 # comment a b: # comment b """) def test_issue_25_01(self) -> None: round_trip("""\ a: # comment 1 # comment 2 - b: # comment 3 c: 1 # comment 4 """) def test_issue_25_02(self) -> None: round_trip("""\ a: # comment 1 # comment 2 - b: 2 # comment 3 """) def test_issue_25_03(self) -> None: s = """\ a: # comment 1 # comment 2 - b: 2 # comment 3 """ round_trip(s, indent=4, block_seq_indent=2) def test_issue_25_04(self) -> None: round_trip("""\ a: # comment 1 # comment 2 b: 1 # comment 3 """) def test_flow_seq_within_seq(self) -> None: round_trip("""\ # comment 1 - a - b # comment 2 - c - d # comment 3 - [e] - f # comment 4 - [] """) def test_comment_after_block_scalar_indicator(self) -> None: round_trip("""\ a: | # abc test 1 test 2 # all done """) test_block_scalar_commented_line_template = """\ y: p # Some comment a: | x {}b: y """ class TestBlockScalarWithComments: # issue 99 reported by Colm O'Connor def test_scalar_with_comments(self) -> None: import ruamel.yaml # NOQA for x in [ "", '\n', '\n# Another comment\n', '\n\n', '\n\n# abc\n#xyz\n', '\n\n# abc\n#xyz\n', '# abc\n\n#xyz\n', '\n\n # abc\n #xyz\n', ]: commented_line = test_block_scalar_commented_line_template.format(x) data = round_trip_load(commented_line) assert round_trip_dump(data) == commented_line