import os import unittest from datetime import datetime as DT from utils.assertions import ( BOLD, CODE, HL, ITALIC, SPAN, STRIKE, UNDERLINED, VERBATIM, WEB_LINK, Doc, Tokens, ) import org_rw from org_rw import MarkerToken, MarkerType, Timestamp, dom, dumps, load, loads DIR = os.path.dirname(os.path.abspath(__file__)) class TestSerde(unittest.TestCase): def test_simple_file_01(self): with open(os.path.join(DIR, "01-simple.org")) as f: doc = load(f) ex = Doc( props=[ ("TITLE", "01-Simple"), ("DESCRIPTION", "Simple org file"), ("TODO", "TODO(t) PAUSED(p) | DONE(d)"), ], children=( HL( "First level", props=[ ("ID", "01-simple-first-level-id"), ("CREATED", DT(2020, 1, 1, 1, 1)), ], content=" First level content\n", children=[ HL( "Second level", props=[("ID", "01-simple-second-level-id")], content="\n Second level content\n", children=[ HL( "Third level", props=[("ID", "01-simple-third-level-id")], content="\n Third level content\n", ) ], ) ], ) ), ) ex.assert_matches(self, doc) def test_mimic_write_file_01(self): """A goal of this library is to be able to update a file without changing parts not directly modified.""" with open(os.path.join(DIR, "01-simple.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(dumps(doc), orig) def test_mimic_write_file_01_second(self): """A goal of this library is to be able to update a file without changing parts not directly modified.""" with open(os.path.join(DIR, "01-simple-2.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(dumps(doc), orig) def test_markup_file_02(self): with open(os.path.join(DIR, "02-markup.org")) as f: doc = load(f) ex = Doc( props=[ ("TITLE", "02-Markup"), ("DESCRIPTION", "Simple org file to test markup"), ("TODO", "TODO(t) PAUSED(p) | DONE(d)"), ], children=( HL( "First level", props=[ ("ID", "02-markup-first-level-id"), ("CREATED", DT(2020, 1, 1, 1, 1)), ], content=[ SPAN(" This is a ", BOLD("bold phrase"), ".\n"), SPAN("\n"), SPAN(" This is a ", VERBATIM("verbatim phrase"), ".\n"), SPAN("\n"), SPAN(" This is a ", ITALIC("italic phrase"), ".\n"), SPAN("\n"), SPAN(" This is a ", STRIKE("strike-through phrase"), ".\n"), SPAN("\n"), SPAN(" This is a ", UNDERLINED("underlined phrase"), ".\n"), SPAN("\n"), SPAN(" This is a ", CODE("code phrase"), ".\n"), SPAN("\n"), SPAN( " This is a nested ", BOLD( [ "bold ", VERBATIM( [ "verbatim ", ITALIC( [ "italic ", STRIKE( [ "strike ", UNDERLINED( [ "underlined ", CODE("code ."), " .", ] ), " .", ] ), " .", ] ), " .", ] ), " .", ] ), ), SPAN("\n"), SPAN("\n"), # THIS IS INTERLEAVED, not nested SPAN( [ " This is a interleaved ", Tokens.BOLD_START, "bold ", Tokens.VERBATIM_START, "verbatim ", Tokens.ITALIC_START, "italic ", Tokens.STRIKE_START, "strike ", Tokens.UNDERLINED_START, "underlined ", Tokens.CODE_START, "code .", Tokens.BOLD_END, " .", Tokens.VERBATIM_END, " .", Tokens.ITALIC_END, " .", Tokens.STRIKE_END, " .", Tokens.UNDERLINED_END, " .", Tokens.CODE_END, "\n", ] ), SPAN("\n"), SPAN( " This is a _ non-underlined phrase because an incorrectly placed content _.\n" ), SPAN("\n"), SPAN( " This is a _ non-underlined phrase because an incorrectly placed content beginning_.\n" ), SPAN("\n"), SPAN(""), SPAN( " This is a _non-underlined phrase because an incorrectly placed content end _.\n" ), SPAN("\n"), SPAN(""), SPAN( " This is a _non-underlined phrase because the lack of an end.\n" ), SPAN("\n"), SPAN("\n"), SPAN( " This is a _non-underlined phrase because an empty line between beginning and\n" ), SPAN("\n"), SPAN(""), SPAN(" end._\n"), ], ) ), ) ex.assert_matches(self, doc) def test_links_file_03(self): with open(os.path.join(DIR, "03-links.org")) as f: doc = load(f) links = list(doc.get_links()) self.assertEqual(len(links), 8) self.assertEqual(links[0].value, "https://codigoparallevar.com/1") self.assertEqual(links[0].description, "web link") self.assertEqual(links[1].value, "https://codigoparallevar.com/2") self.assertEqual(links[1].description, "web link") self.assertEqual(links[2].value, "* First level") self.assertEqual(links[2].description, None) self.assertEqual(links[3].value, "id:03-markup-first-level-id") self.assertEqual(links[3].description, "a link to a section by id") self.assertEqual(links[4].value, "https://codigoparallevar.com/3") self.assertEqual(links[4].description, "web link") self.assertEqual(links[5].value, "https://codigoparallevar.com/4") self.assertEqual(links[5].description, "[tricky web link]\u200b") self.assertEqual(links[6].value, "https://codigoparallevar.com/5") self.assertEqual(links[6].description, "another tricky web link") self.assertEqual(links[7].value, "https://codigoparallevar.com/implicit") self.assertEqual(links[7].description, "https://codigoparallevar.com/implicit") ex = Doc( props=[ ("TITLE", "03-Links"), ("DESCRIPTION", "Simple org file to test links"), ("TODO", "TODO(t) PAUSED(p) | DONE(d)"), ], children=( HL( "First level", props=[ ("ID", "03-markup-first-level-id"), ("CREATED", DT(2020, 1, 1, 1, 1)), ], content=[ SPAN( " This is a ", WEB_LINK("web link", "https://codigoparallevar.com/1"), ".\n", ), SPAN("\n"), SPAN( " This is a ", ITALIC( [ "italized ", WEB_LINK( "web link", "https://codigoparallevar.com/2" ), ] ), ".\n", ), SPAN("\n"), SPAN( " This is a link with no description to ", WEB_LINK(None, "* First level"), ".\n", ), SPAN("\n"), SPAN( " This is ", WEB_LINK( "a link to a section by id", "id:03-markup-first-level-id", ), ".\n", ), SPAN("\n"), SPAN( " This is a ", WEB_LINK("web link", "https://codigoparallevar.com/3"), " followed up with some text.\n", ), SPAN("\n"), SPAN( " This is a ", WEB_LINK( "[tricky web link]\u200b", "https://codigoparallevar.com/4", ), " followed up with some text.\n", ), SPAN("\n"), SPAN( " This is [", WEB_LINK( "another tricky web link", "https://codigoparallevar.com/5", ), "] followed up with some text.\n", ), ], ), HL( "Implicit links", props=[ ("ID", "03-markup-implicit-links"), ("CREATED", DT(2020, 1, 1, 1, 1)), ], content=[ SPAN( " This is an implicit web link: https://codigoparallevar.com/implicit.\n", ), ], ), ), ) ex.assert_matches(self, doc) def test_update_links_file_03(self): with open(os.path.join(DIR, "03-links.org")) as f: doc = load(f) links = list(doc.get_links()) self.assertEqual(len(links), 8) self.assertEqual(links[0].value, "https://codigoparallevar.com/1") self.assertEqual(links[0].description, "web link") links[0].value = "https://codigoparallevar.com/1-updated" links[0].description = "web link #1 with update" self.assertEqual(links[1].value, "https://codigoparallevar.com/2") self.assertEqual(links[1].description, "web link") links[1].value = "https://codigoparallevar.com/2-updated" links[1].description = "web link #2 with update" self.assertEqual(links[2].value, "* First level") self.assertEqual(links[2].description, None) links[2].value = "* Non-existent level" links[2].description = "a description now" self.assertEqual(links[3].value, "id:03-markup-first-level-id") self.assertEqual(links[3].description, "a link to a section by id") links[3].value = "id:03-markup-non-existent-level-id" links[3].description = None self.assertEqual(links[4].value, "https://codigoparallevar.com/3") self.assertEqual(links[4].description, "web link") links[4].value = "https://codigoparallevar.com/3-updated" links[4].description = "web link #3 with update" self.assertEqual(links[5].value, "https://codigoparallevar.com/4") self.assertEqual(links[5].description, "[tricky web link]\u200b") links[5].value = "https://codigoparallevar.com/4-updated" links[5].description = "[tricky web link #4 with update]" self.assertEqual(links[6].value, "https://codigoparallevar.com/5") self.assertEqual(links[6].description, "another tricky web link") links[6].value = "https://codigoparallevar.com/5-updated" links[6].description = "another tricky web link #5 with update" self.assertEqual(links[7].value, "https://codigoparallevar.com/implicit") self.assertEqual(links[7].description, "https://codigoparallevar.com/implicit") ex = Doc( props=[ ("TITLE", "03-Links"), ("DESCRIPTION", "Simple org file to test links"), ("TODO", "TODO(t) PAUSED(p) | DONE(d)"), ], children=( HL( "First level", props=[ ("ID", "03-markup-first-level-id"), ("CREATED", DT(2020, 1, 1, 1, 1)), ], content=[ SPAN( " This is a ", WEB_LINK( "web link #1 with update", "https://codigoparallevar.com/1-updated", ), ".\n", ), SPAN("\n"), SPAN( " This is a ", ITALIC( [ "italized ", WEB_LINK( "web link #2 with update", "https://codigoparallevar.com/2-updated", ), ] ), ".\n", ), SPAN("\n"), SPAN( " This is a link with no description to ", WEB_LINK("a description now", "* Non-existent level"), ".\n", ), SPAN("\n"), SPAN( " This is ", WEB_LINK( None, "id:03-markup-non-existent-level-id", ), ".\n", ), SPAN("\n"), SPAN( " This is a ", WEB_LINK( "web link #3 with update", "https://codigoparallevar.com/3-updated", ), " followed up with some text.\n", ), SPAN("\n"), SPAN( " This is a ", WEB_LINK( "[tricky web link #4 with update]", "https://codigoparallevar.com/4-updated", ), " followed up with some text.\n", ), SPAN("\n"), SPAN( " This is [", WEB_LINK( "another tricky web link #5 with update", "https://codigoparallevar.com/5-updated", ), "] followed up with some text.\n", ), ], ), HL( "Implicit links", props=[ ("ID", "03-markup-implicit-links"), ("CREATED", DT(2020, 1, 1, 1, 1)), ], content=[ SPAN( " This is an implicit web link: https://codigoparallevar.com/implicit.\n", ), ], ), ), ) ex.assert_matches(self, doc) def test_mimic_write_file_04(self): with open(os.path.join(DIR, "04-code.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(dumps(doc), orig) def test_code_file_04(self): with open(os.path.join(DIR, "04-code.org")) as f: doc = load(f) snippets = list(doc.get_code_snippets()) self.assertEqual(len(snippets), 3) self.assertEqual(snippets[0].name, "first-code-name") self.assertEqual(snippets[0].language, "shell") self.assertEqual( snippets[0].content, 'echo "This is a test"\n' + 'echo "with two lines"\n' + "exit 0 # Exit successfully", ) self.assertEqual(snippets[0].arguments.split(), [":results", "verbatim"]) self.assertEqual( snippets[0].result, "This is a test\n" + "with two lines", ) self.assertEqual(snippets[1].name, None) self.assertEqual(snippets[1].language, "shell") self.assertEqual( snippets[1].content, 'echo "This is another test"\n' + 'echo "with two lines too"\n' + "exit 0 # Comment", ) self.assertEqual( snippets[1].result, "This is another test\n" + "with two lines too" ) self.assertEqual(snippets[2].name, None) self.assertEqual(snippets[2].language, "c") self.assertEqual( snippets[2].content, "/* This code has to be escaped to\n" + " * avoid confusion with new headlines.\n" + " */\n" + "main(){}", ) def test_mimic_write_file_05(self): with open(os.path.join(DIR, "05-dates.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(dumps(doc), orig) def test_planning_info_file_05(self): with open(os.path.join(DIR, "05-dates.org")) as f: orig = f.read() doc = loads(orig) hl = doc.getTopHeadlines()[0] self.assertEqual( hl.scheduled.time, Timestamp(True, 2020, 12, 12, "Sáb", None, None) ) self.assertEqual( hl.closed.time, Timestamp(True, 2020, 12, 13, "Dom", None, None) ) self.assertEqual( hl.deadline.time, Timestamp(True, 2020, 12, 14, "Lun", None, None) ) hl_schedule_range = hl.children[0] self.assertEqual( hl_schedule_range.scheduled.time, Timestamp(True, 2020, 12, 15, "Mar", 0, 5) ) self.assertEqual( hl_schedule_range.scheduled.end_time, Timestamp(True, 2020, 12, 15, "Mar", 0, 10), ) hl_schedule_range = hl.children[1] self.assertEqual( hl_schedule_range.scheduled.time, Timestamp(True, 2020, 12, 15, "Mar", 0, 5, "++1w"), ) self.assertEqual( hl_schedule_range.scheduled.end_time, Timestamp(True, 2020, 12, 15, "Mar", 0, 10), ) self.assertEqual( hl_schedule_range.scheduled.repetition, "++1w", ) def test_update_info_file_05(self): with open(os.path.join(DIR, "05-dates.org")) as f: orig = f.read() doc = loads(orig) hl = doc.getTopHeadlines()[0] hl.scheduled.time.day = 15 hl.closed.time.day = 16 hl.deadline.time.day = 17 # Account for removeing 3 days-of-week + 1 space each self.assertEqual(len(dumps(doc)), len(orig) - (4) * 3) doc_updated = loads(dumps(doc)) hl_up = doc_updated.getTopHeadlines()[0] self.assertEqual( hl.scheduled.time, Timestamp(True, 2020, 12, 15, None, None, None) ) self.assertEqual( hl.closed.time, Timestamp(True, 2020, 12, 16, None, None, None) ) self.assertEqual( hl.deadline.time, Timestamp(True, 2020, 12, 17, None, None, None) ) def test_mimic_write_file_06(self): with open(os.path.join(DIR, "06-lists.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(dumps(doc), orig) def test_structure_file_06(self): with open(os.path.join(DIR, "06-lists.org")) as f: orig = f.read() doc = loads(orig) hl = doc.getTopHeadlines()[0] # ... lists = hl.getLists() self.assertEqual(len(lists), 3) self.assertEqual(lists[0][0].content, ["This is a simple list."]) self.assertEqual(lists[0][0].bullet, "-") self.assertEqual( lists[0][1].content, [ "This list has multiple elements, with ", MarkerToken(closing=False, tok_type=MarkerType.UNDERLINED_MODE), "markup", MarkerToken(closing=True, tok_type=MarkerType.UNDERLINED_MODE), ".", "\n", ], ) self.assertEqual(lists[1][0].content, ["This is a simple list."]) self.assertEqual(lists[1][0].bullet, "+") hl2 = doc.getTopHeadlines()[1] # ... lists2 = hl2.getLists() self.assertEqual(len(lists2), 2) self.assertEqual(lists2[0][0].content, ["First element"]) self.assertEqual(lists2[0][0].counter, "1") self.assertEqual(lists2[0][0].counter_sep, ".") self.assertEqual(lists2[0][1].content, ["Second element", "\n"]) self.assertEqual(lists2[0][1].counter, "2") self.assertEqual(lists2[0][1].counter_sep, ".") self.assertEqual(lists2[1][0].content, ["First element"]) self.assertEqual(lists2[1][0].counter, "1") self.assertEqual(lists2[1][0].counter_sep, ")") self.assertEqual(lists2[1][1].content, ["Second element", "\n"]) self.assertEqual(lists2[1][1].counter, "2") self.assertEqual(lists2[1][1].counter_sep, ")") hl4 = doc.getTopHeadlines()[3] # ... lists4 = hl4.getLists() print(lists4) self.assertEqual(len(lists4), 2) self.assertEqual( lists4[0][0].content, ["This is a list item...", "\n that spans multiple lines", "\n"], ) self.assertEqual(lists4[0][0].bullet, "-") self.assertEqual( lists4[0][1].content, [ "This is another list item...", "\n that has content on multiple lines", "\n", ], ) self.assertEqual(lists4[0][1].bullet, "-") self.assertEqual( lists4[1][0].content, ["This is another", "\n multiline list", "\n"] ) self.assertEqual(lists4[1][0].bullet, "-") def test_org_roam_07(self): with open(os.path.join(DIR, "07-org-roam-v2.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(doc.get_property("ID"), "515054a9-ced8-4119-a844-71726f80dedf") self.assertEqual(len(doc.getTopHeadlines()), 1) hl = doc.getTopHeadlines()[0] self.assertEqual(hl.get_property("ID"), "419f4651-21c8-4166-b8d5-692c34be9f93") self.assertEqual(len(hl.children), 1) def test_org_property_creation_08(self): with open(os.path.join(DIR, "08-property-creation.org")) as f: orig = f.read() doc = loads(orig) headline = doc.getTopHeadlines()[0] headline.id = "first" second = headline.children[0] second.id = "second" self.assertEqual( dumps(doc).strip(), """ #+TITLE: 08-Property-creation * Top headline :PROPERTIES: :ID: first :END: ** Second headline :PROPERTIES: :ID: second :END: *** Third headline """.strip(), ) def test_markup_file_09(self): with open(os.path.join(DIR, "09-markup-on-headline.org")) as f: doc = load(f) hl = doc.getTopHeadlines()[0] print(hl.title) self.assertEqual( hl.title.contents, [ "Headline ", MarkerToken(closing=False, tok_type=MarkerType.UNDERLINED_MODE), "with", MarkerToken(closing=True, tok_type=MarkerType.UNDERLINED_MODE), " markup", ], ) def test_mimic_write_file_10(self): with open(os.path.join(DIR, "10-tables.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(dumps(doc), orig) def test_tables_file_10(self): with open(os.path.join(DIR, "10-tables.org")) as f: doc = load(f) hl = doc.getTopHeadlines()[0] tables = hl.get_tables() first_table = tables[0] self.assertEqual(len(first_table), 4) print(first_table[0]) self.assertEqual(len(first_table[0].cells), 3) self.assertEqual(first_table[0].cells[0].strip(), "Header1") self.assertEqual(first_table[0].cells[1].strip(), "Header2") self.assertEqual(first_table[0].cells[2].strip(), "Header3") hl = hl.children[0] tables = hl.get_tables() first_table = tables[0] self.assertEqual(len(first_table), 4) print(first_table[0]) self.assertEqual(len(first_table[0].cells), 3) self.assertEqual(first_table[0].cells[0].strip(), "Header1") self.assertEqual(first_table[0].cells[1].strip(), "Header2") self.assertEqual(first_table[0].cells[2].strip(), "Header3") def test_tables_html_file_10(self): with open(os.path.join(DIR, "10-tables.org")) as f: doc = load(f) hl = doc.getTopHeadlines()[0] tree = hl.as_dom() non_props = [ item for item in tree if not isinstance(item, dom.PropertyDrawerNode) ] self.assertTrue( isinstance(non_props[0], dom.Text) and isinstance(non_props[1], dom.TableNode) and isinstance(non_props[2], dom.Text), "Expected ", ) hl = hl.children[0] tree = hl.as_dom() non_props = [ item for item in tree if not ( isinstance(item, dom.PropertyDrawerNode) or isinstance(item, dom.Text) ) ] print_tree(non_props) self.assertTrue(len(non_props) == 1, "Expected , with only (1) element") def test_nested_lists_html_file_11(self): with open(os.path.join(DIR, "11-nested-lists.org")) as f: doc = load(f) hl = doc.getTopHeadlines()[0] tree = hl.as_dom() non_props = [ item for item in tree if not isinstance(item, dom.PropertyDrawerNode) ] print_tree(non_props) self.assertTrue( (len(non_props) == 1) and (isinstance(non_props[0], dom.ListGroupNode)), "Expected only as top level", ) dom_list = non_props[0] children = dom_list.children self.assertTrue( len(children) == 5, "Expected 5 items inside , 3 texts and 2 sublists" ) # Assert texts self.assertEqual(children[0].content, ["1"]) self.assertEqual(children[2].content, ["2"]) self.assertEqual( children[4].content[0], "3" ) # Might be ['3', '\n'] but shouldn't be a breaking change # Assert lists self.assertTrue( isinstance(children[1], dom.ListGroupNode), 'Expected sublist inside "1"' ) self.assertEqual(children[1].children[0].content, ["1.1"]) self.assertEqual(children[1].children[1].content, ["1.2"]) self.assertTrue( isinstance(children[3], dom.ListGroupNode), 'Expected sublist inside "2"' ) self.assertEqual(children[3].children[0].content, ["2.1"]) self.assertEqual(children[3].children[1].content, ["2.2"]) def test_mimic_write_file_12(self): with open(os.path.join(DIR, "12-headlines-with-skip-levels.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(dumps(doc), orig) def test_add_todo_keywords_programatically(self): orig = """* NEW_TODO_STATE First entry * NEW_DONE_STATE Second entry""" doc = loads( orig, environment={"org-todo-keywords": "NEW_TODO_STATE | NEW_DONE_STATE"} ) self.assertEqual(doc.headlines[0].is_todo, True) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[1].is_todo, False) self.assertEqual(doc.headlines[1].is_done, True) self.assertEqual(dumps(doc), orig) def test_add_todo_keywords_in_file(self): orig = """#+TODO: NEW_TODO_STATE | NEW_DONE_STATE * NEW_TODO_STATE First entry * NEW_DONE_STATE Second entry""" doc = loads( orig, environment={"org-todo-keywords": "NEW_TODO_STATE | NEW_DONE_STATE"} ) self.assertEqual(doc.headlines[0].is_todo, True) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[1].is_todo, False) self.assertEqual(doc.headlines[1].is_done, True) self.assertEqual(dumps(doc), orig) def test_mimic_write_file_13(self): with open(os.path.join(DIR, "13-tags.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(dumps(doc), orig) def test_tag_property_read_13(self): with open(os.path.join(DIR, "13-tags.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(doc.tags, ["filetag"]) h1_1, h1_2 = doc.getTopHeadlines() self.assertEqual(sorted(h1_1.tags), ["filetag", "h1tag"]) self.assertEqual(sorted(h1_2.tags), ["filetag", "otherh1tag"]) h1_1_h2 = h1_1.children[0] self.assertEqual(sorted(h1_1_h2.tags), ["filetag", "h1tag", "h2tag"]) h1_2_h2 = h1_2.children[0] self.assertEqual(sorted(h1_2_h2.tags), ["filetag", "otherh1tag", "otherh2tag"]) def test_shallow_tag_property_read_13(self): with open(os.path.join(DIR, "13-tags.org")) as f: orig = f.read() doc = loads(orig) self.assertEqual(doc.shallow_tags, ["filetag"]) h1_1, h1_2 = doc.getTopHeadlines() self.assertEqual(sorted(h1_1.shallow_tags), ["h1tag"]) self.assertEqual(sorted(h1_2.shallow_tags), ["otherh1tag"]) h1_1_h2 = h1_1.children[0] self.assertEqual(sorted(h1_1_h2.shallow_tags), ["h2tag"]) h1_2_h2 = h1_2.children[0] self.assertEqual(sorted(h1_2_h2.shallow_tags), ["otherh2tag"]) def test_exclude_tags_from_inheritance_property_read_13(self): with open(os.path.join(DIR, "13-tags.org")) as f: orig = f.read() doc = loads( orig, { "org-tags-exclude-from-inheritance": ("h1tag", "otherh2tag"), }, ) self.assertEqual(doc.tags, ["filetag"]) h1_1, h1_2 = doc.getTopHeadlines() self.assertEqual(sorted(h1_1.tags), ["filetag", "h1tag"]) self.assertEqual(sorted(h1_2.tags), ["filetag", "otherh1tag"]) h1_1_h2 = h1_1.children[0] self.assertEqual(sorted(h1_1_h2.tags), ["filetag", "h2tag"]) h1_2_h2 = h1_2.children[0] self.assertEqual(sorted(h1_2_h2.tags), ["filetag", "otherh1tag", "otherh2tag"]) def test_select_tags_to_inheritance_property_read_13(self): with open(os.path.join(DIR, "13-tags.org")) as f: orig = f.read() doc = loads( orig, { "org-tags-exclude-from-inheritance": ("h1tag", "otherh2tag"), "org-use-tag-inheritance": ("h1tag",), }, ) self.assertEqual(doc.tags, ["filetag"]) h1_1, h1_2 = doc.getTopHeadlines() self.assertEqual(sorted(h1_1.tags), ["h1tag"]) self.assertEqual(sorted(h1_2.tags), ["otherh1tag"]) h1_1_h2 = h1_1.children[0] self.assertEqual(sorted(h1_1_h2.tags), ["h1tag", "h2tag"]) h1_2_h2 = h1_2.children[0] self.assertEqual(sorted(h1_2_h2.tags), ["otherh2tag"]) def test_update_headline_from_none_to_todo(self): orig = "* First entry" doc = loads(orig) self.assertEqual(doc.headlines[0].is_todo, False) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[0].state, None) doc.headlines[0].state = "TODO" self.assertEqual(doc.headlines[0].is_todo, True) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[0].state["name"], "TODO") self.assertEqual(dumps(doc), "* TODO First entry") def test_update_headline_from_none_to_done(self): orig = "* First entry" doc = loads(orig) self.assertEqual(doc.headlines[0].is_todo, False) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[0].state, None) doc.headlines[0].state = org_rw.HeadlineState(name="DONE") self.assertEqual(doc.headlines[0].is_todo, False) self.assertEqual(doc.headlines[0].is_done, True) self.assertEqual(doc.headlines[0].state["name"], "DONE") self.assertEqual(dumps(doc), "* DONE First entry") def test_update_headline_from_todo_to_none(self): orig = "* TODO First entry" doc = loads(orig) self.assertEqual(doc.headlines[0].is_todo, True) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[0].state["name"], "TODO") doc.headlines[0].state = None self.assertEqual(doc.headlines[0].is_todo, False) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[0].state, None) self.assertEqual(dumps(doc), "* First entry") def test_update_headline_from_todo_to_done(self): orig = "* TODO First entry" doc = loads(orig) self.assertEqual(doc.headlines[0].is_todo, True) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[0].state["name"], "TODO") doc.headlines[0].state = "DONE" self.assertEqual(doc.headlines[0].is_todo, False) self.assertEqual(doc.headlines[0].is_done, True) self.assertEqual(doc.headlines[0].state["name"], "DONE") self.assertEqual(dumps(doc), "* DONE First entry") def test_update_headline_from_done_to_todo(self): orig = "* DONE First entry" doc = loads(orig) self.assertEqual(doc.headlines[0].is_todo, False) self.assertEqual(doc.headlines[0].is_done, True) self.assertEqual(doc.headlines[0].state["name"], "DONE") doc.headlines[0].state = org_rw.HeadlineState(name="TODO") self.assertEqual(doc.headlines[0].is_todo, True) self.assertEqual(doc.headlines[0].is_done, False) self.assertEqual(doc.headlines[0].state["name"], "TODO") self.assertEqual(dumps(doc), "* TODO First entry") def print_tree(tree, indentation=0, headline=None): for element in tree: print(" " * indentation * 2, "EL:", element) if "children" in dir(element): if len(element.children) > 0: print_element(element.children, indentation + 1, headline) print() elif "content" in dir(element): for content in element.content: print_element(content, indentation + 1, headline) def print_element(element, indentation, headline): if isinstance(element, org_rw.Link): print(" " * indentation * 2, "Link:", element.get_raw()) elif isinstance(element, str): print( " " * indentation * 2, "Str[" + element.replace("\n", "") + "]", type(element), ) else: print_tree(element, indentation, headline)