1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00
ruby--ruby/test/rdoc/test_rdoc_markup_parser.rb
drbrain 1c279a7d27 * lib/rdoc*: Updated to RDoc 4.0 (pre-release)
* bin/rdoc:  ditto
* test/rdoc:  ditto
* NEWS:  Updated with RDoc 4.0 information


git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@37889 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2012-11-27 04:28:14 +00:00

1685 lines
32 KiB
Ruby

# coding: utf-8
require 'rdoc/test_case'
class TestRDocMarkupParser < RDoc::TestCase
def setup
super
@RMP = @RM::Parser
end
def mu_pp(obj)
s = ''
s = PP.pp obj, s
s = s.force_encoding(Encoding.default_external) if defined? Encoding
s.chomp
end
def test_build_heading
parser = @RMP.new
parser.tokens.replace [
[:TEXT, 'heading three', 4, 0],
[:NEWLINE, "\n", 17, 0],
]
assert_equal @RM::Heading.new(3, 'heading three'), parser.build_heading(3)
end
def test_char_pos
parser = @RMP.new
s = parser.setup_scanner 'cät'
s.scan(/\S+/)
if Object.const_defined? :Encoding then
assert_equal 3, parser.char_pos(s.pos)
else
assert_equal 4, parser.char_pos(s.pos)
end
end
def test_get
parser = util_parser
assert_equal [:HEADER, 1, 0, 0], parser.get
assert_equal 7, parser.tokens.length
end
def test_parse_bullet
str = <<-STR
* l1
* l2
STR
expected = [
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1')),
@RM::ListItem.new(nil,
@RM::Paragraph.new('l2'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_bullet_utf_8
str = <<-STR
*
STR
expected = [
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('新しい機能'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_bullet_verbatim_heading
str = <<-STR
* l1
v
= H
STR
expected = [
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1'),
@RM::Verbatim.new("v\n"))]),
@RM::Heading.new(1, 'H')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_bullet_heading
str = <<-STR
* = l1
STR
expected = [
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Heading.new(1, 'l1'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_bullet_indent
str = <<-STR
* l1
* l1.1
* l2
STR
expected = [
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1'),
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1.1'))])),
@RM::ListItem.new(nil,
@RM::Paragraph.new('l2'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_bullet_paragraph
str = <<-STR
now is
* l1
* l2
the time
STR
expected = [
@RM::Paragraph.new('now is'),
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1')),
@RM::ListItem.new(nil,
@RM::Paragraph.new('l2')),
]),
@RM::Paragraph.new('the time'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_bullet_multiline
str = <<-STR
* l1
l1+
* l2
STR
expected = [
list(:BULLET,
item(nil,
para('l1 ', 'l1+')),
item(nil,
para('l2')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_bullet_multiparagraph
str = <<-STR
* l1
l1+
STR
expected = [
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1'),
@RM::BlankLine.new,
@RM::Paragraph.new('l1+')),
]),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_bullet_indent_verbatim
str = <<-STR
* l1
* l1.1
text
code
code
text
* l2
STR
expected = [
list(:BULLET,
item(nil,
para('l1'),
list(:BULLET,
item(nil,
para('l1.1 ', 'text'),
verb("code\n", " code\n"),
para('text')))),
item(nil,
para('l2')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_dash
str = <<-STR
- one
- two
STR
expected = [
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('one')),
@RM::ListItem.new(nil,
@RM::Paragraph.new('two'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading
str = '= heading one'
expected = [
@RM::Heading.new(1, 'heading one')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_three
str = '=== heading three'
expected = [
@RM::Heading.new(3, 'heading three')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_bullet
str = '= * heading one'
expected = [
@RM::Heading.new(1, '* heading one')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_empty
str = <<-STR
===
* bullet
STR
expected = [
@RM::Heading.new(3, ''),
@RM::BlankLine.new,
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('bullet'))]),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_heading
str = '= ='
expected = [
@RM::Heading.new(1, '=')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_lalpha
str = '= b. heading one'
expected = [
@RM::Heading.new(1, 'b. heading one')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_label
str = '= [heading one]'
expected = [
@RM::Heading.new(1, '[heading one]')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_note
str = '= heading one::'
expected = [
@RM::Heading.new(1, 'heading one::')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_number
str = '= 5. heading one'
expected = [
@RM::Heading.new(1, '5. heading one')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_heading_ualpha
str = '= B. heading one'
expected = [
@RM::Heading.new(1, 'B. heading one')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_label
str = <<-STR
[one] item one
[two] item two
STR
expected = [
list(:LABEL,
item(%w[one],
para('item one')),
item(%w[two],
para('item two')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_label_bullet
str = <<-STR
[cat] l1
* l1.1
[dog] l2
STR
expected = [
list(:LABEL,
item(%w[cat],
para('l1'),
list(:BULLET,
item(nil,
para('l1.1')))),
item(%w[dog],
para('l2')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_label_multi_label
str = <<-STR
[one]
[two] some description
STR
expected = [
list(:LABEL,
item(%w[one two],
para('some description')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_label_multi_line
str = <<-STR
[cat] l1
continuation
[dog] l2
STR
expected = [
list(:LABEL,
item(%w[cat],
para('l1 ', 'continuation')),
item(%w[dog],
para('l2')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_label_newline
str = <<-STR
[one]
item one
[two]
item two
STR
expected = [
list(:LABEL,
item(%w[one],
para('item one')),
item(%w[two],
para('item two')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_lalpha
str = <<-STR
a. l1
b. l2
STR
expected = [
@RM::List.new(:LALPHA, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1')),
@RM::ListItem.new(nil,
@RM::Paragraph.new('l2'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_lalpha_ualpha
str = <<-STR
a. l1
b. l2
A. l3
A. l4
STR
expected = [
@RM::List.new(:LALPHA, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1')),
@RM::ListItem.new(nil,
@RM::Paragraph.new('l2'))]),
@RM::List.new(:UALPHA, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l3')),
@RM::ListItem.new(nil,
@RM::Paragraph.new('l4'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_lalpha_utf_8
str = <<-STR
a.
STR
expected = [
@RM::List.new(:LALPHA, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('新しい機能'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_line_break
str = "now is\nthe time \nfor all"
expected = [
para('now is ', 'the time'),
blank_line,
para('for all')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_list_list_1
str = <<-STR
10. para 1
[label 1]
para 1.1
code
para 1.2
STR
expected = [
list(:NUMBER,
item(nil,
para('para 1'),
blank_line,
list(:LABEL,
item(%w[label\ 1],
para('para 1.1'),
blank_line,
verb("code\n"),
para('para 1.2')))))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_list_list_2
str = <<-STR
6. para
label 1:: text 1
label 2:: text 2
STR
expected = [
list(:NUMBER,
item(nil,
para('para'),
blank_line,
list(:NOTE,
item(%w[label\ 1],
para('text 1')),
item(%w[label\ 2],
para('text 2')))))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_list_verbatim
str = <<-STR
* one
verb1
verb2
* two
STR
expected = [
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('one'),
@RM::Verbatim.new("verb1\n", "verb2\n")),
@RM::ListItem.new(nil,
@RM::Paragraph.new('two'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_lists
str = <<-STR
now is
* l1
1. n1
2. n2
* l2
the time
STR
expected = [
@RM::Paragraph.new('now is'),
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1'))]),
@RM::List.new(:NUMBER, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('n1')),
@RM::ListItem.new(nil,
@RM::Paragraph.new('n2'))]),
@RM::List.new(:BULLET, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l2'))]),
@RM::Paragraph.new('the time')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_note
str = <<-STR
one:: item one
two:: item two
STR
expected = [
list(:NOTE,
item(%w[one],
para('item one')),
item(%w[two],
para('item two')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_note_empty
str = <<-STR
one::
two::
STR
expected = [
list(:NOTE,
item(%w[one two],
blank_line))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_note_note
str = <<-STR
one:: two::
STR
expected = [
list(:NOTE,
item(%w[one],
list(:NOTE,
item(%w[two],
blank_line))))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_number_bullet
str = <<-STR
1. l1
* l1.1
2. l2
STR
expected = [
list(:NUMBER,
item(nil,
para('l1'),
list(:BULLET,
item(nil,
para('l1.1')))),
item(nil,
para('l2')))]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_paragraph
str = <<-STR
now is the time
for all good men
STR
expected = [
@RM::Paragraph.new('now is the time'),
@RM::BlankLine.new,
@RM::Paragraph.new('for all good men')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_paragraph_multiline
str = "now is the time\nfor all good men"
expected = @RM::Paragraph.new 'now is the time ', 'for all good men'
assert_equal [expected], @RMP.parse(str).parts
end
def test_parse_paragraph_verbatim
str = <<-STR
now is the time
code _line_ here
for all good men
STR
expected = [
@RM::Paragraph.new('now is the time'),
@RM::Verbatim.new("code _line_ here\n"),
@RM::Paragraph.new('for all good men'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_rule
str = <<-STR
now is the time
---
for all good men
STR
expected = [
@RM::Paragraph.new('now is the time'),
@RM::BlankLine.new,
@RM::Rule.new(1),
@RM::BlankLine.new,
@RM::Paragraph.new('for all good men')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_ualpha
str = <<-STR
A. l1
B. l2
STR
expected = [
@RM::List.new(:UALPHA, *[
@RM::ListItem.new(nil,
@RM::Paragraph.new('l1')),
@RM::ListItem.new(nil,
@RM::Paragraph.new('l2'))])]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_trailing_cr
expected = [ @RM::Paragraph.new('Text') ]
# FIXME hangs the parser:
assert_equal expected, @RMP.parse("Text\r").parts
end
def test_parse_verbatim
str = <<-STR
now is
code
the time
STR
expected = [
@RM::Paragraph.new('now is'),
@RM::Verbatim.new("code\n"),
@RM::Paragraph.new('the time'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_bullet
str = <<-STR
* blah
STR
expected = [
@RM::Verbatim.new("* blah\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_dash
str = <<-STR
- blah
STR
expected = [
@RM::Verbatim.new("- blah\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_fold
str = <<-STR
now is
code
code1
the time
STR
expected = [
@RM::Paragraph.new('now is'),
@RM::Verbatim.new("code\n", "\n", "code1\n"),
@RM::Paragraph.new('the time'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_heading
str = <<-STR
text
=== heading three
STR
expected = [
@RM::Paragraph.new('text'),
@RM::Verbatim.new("=== heading three\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_heading2
str = "text\n code\n=== heading three"
expected = [
@RM::Paragraph.new('text'),
@RM::Verbatim.new("code\n"),
@RM::Heading.new(3, 'heading three')]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_label
str = <<-STR
[blah] blah
STR
expected = [
@RM::Verbatim.new("[blah] blah\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_lalpha
str = <<-STR
b. blah
STR
expected = [
@RM::Verbatim.new("b. blah\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_markup_example
str = <<-STR
text
code
=== heading three
STR
expected = [
@RM::Paragraph.new('text'),
@RM::Verbatim.new("code\n", "=== heading three\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_merge
str = <<-STR
now is
code
code1
the time
STR
expected = [
@RM::Paragraph.new('now is'),
@RM::Verbatim.new("code\n", "\n", "code1\n"),
@RM::Paragraph.new('the time'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_merge2
str = <<-STR
now is
code
code1
code2
the time
STR
expected = [
@RM::Paragraph.new('now is'),
@RM::Verbatim.new("code\n", "\n", "code1\n", "\n", "code2\n"),
@RM::Paragraph.new('the time'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_multiline
str = <<-STR
now is
code
code1
the time
STR
expected = [
@RM::Paragraph.new('now is'),
@RM::Verbatim.new("code\n", "code1\n"),
@RM::Paragraph.new('the time'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_multilevel
str = <<-STR
now is the time
code
more code
for all good men
STR
expected = [
@RM::Paragraph.new('now is the time'),
@RM::Verbatim.new(" code\n",
"more code\n"),
@RM::Paragraph.new('for all good men'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_note
str = <<-STR
blah:: blah
STR
expected = [
@RM::Verbatim.new("blah:: blah\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_number
str = <<-STR
2. blah
STR
expected = [
@RM::Verbatim.new("2. blah\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_rule
str = <<-STR
text
--- lib/blah.rb.orig
+++ lib/blah.rb
STR
expected = [
@RM::Paragraph.new('text'),
@RM::BlankLine.new,
@RM::Verbatim.new("--- lib/blah.rb.orig\n",
"+++ lib/blah.rb\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_rule2
str = <<-STR.chomp
text
---
STR
expected = [
@RM::Paragraph.new('text'),
@RM::BlankLine.new,
@RM::Verbatim.new("---")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_trim
str = <<-STR
now is
code
code1
the time
STR
expected = [
@RM::Paragraph.new('now is'),
@RM::Verbatim.new("code\n",
"\n",
"code1\n"),
@RM::Paragraph.new('the time'),
]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_verbatim_ualpha
str = <<-STR
B. blah
STR
expected = [
@RM::Verbatim.new("B. blah\n")]
assert_equal expected, @RMP.parse(str).parts
end
def test_parse_whitespace
expected = [
@RM::Paragraph.new('hello'),
]
assert_equal expected, @RMP.parse('hello').parts
expected = [
@RM::Verbatim.new('hello '),
]
assert_equal expected, @RMP.parse(' hello ').parts
expected = [
@RM::Verbatim.new('hello '),
]
assert_equal expected, @RMP.parse(' hello ').parts
expected = [
@RM::Paragraph.new('1'),
@RM::Verbatim.new("2\n", ' 3'),
]
assert_equal expected, @RMP.parse("1\n 2\n 3").parts
expected = [
@RM::Verbatim.new("1\n",
" 2\n",
" 3"),
]
assert_equal expected, @RMP.parse(" 1\n 2\n 3").parts
expected = [
@RM::Paragraph.new('1'),
@RM::Verbatim.new("2\n",
" 3\n"),
@RM::Paragraph.new('1'),
@RM::Verbatim.new('2'),
]
assert_equal expected, @RMP.parse("1\n 2\n 3\n1\n 2").parts
expected = [
@RM::Verbatim.new("1\n",
" 2\n",
" 3\n",
"1\n",
' 2'),
]
assert_equal expected, @RMP.parse(" 1\n 2\n 3\n 1\n 2").parts
expected = [
@RM::Verbatim.new("1\n",
" 2\n",
"\n",
' 3'),
]
assert_equal expected, @RMP.parse(" 1\n 2\n\n 3").parts
end
def test_peek_token
parser = util_parser
assert_equal [:HEADER, 1, 0, 0], parser.peek_token
assert_equal 8, parser.tokens.length
end
def test_skip
parser = util_parser
assert_equal [:HEADER, 1, 0, 0], parser.skip(:HEADER)
assert_equal [:TEXT, 'Heading', 2, 0], parser.get
assert_equal [:NEWLINE, "\n", 9, 0], parser.peek_token
assert_raises RDoc::Markup::Parser::ParseError do
parser.skip :NONE
end
assert_equal [:NEWLINE, "\n", 9, 0], parser.peek_token
assert_equal nil, parser.skip(:NONE, false)
assert_equal [:NEWLINE, "\n", 9, 0], parser.peek_token
end
def test_tokenize_bullet
str = <<-STR
* l1
STR
expected = [
[:BULLET, '*', 0, 0],
[:TEXT, 'l1', 2, 0],
[:NEWLINE, "\n", 4, 0],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_bullet_indent
str = <<-STR
* l1
* l1.1
STR
expected = [
[:BULLET, '*', 0, 0],
[:TEXT, 'l1', 2, 0],
[:NEWLINE, "\n", 4, 0],
[:BULLET, '*', 2, 1],
[:TEXT, 'l1.1', 4, 1],
[:NEWLINE, "\n", 8, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_heading
str = <<-STR
= Heading
== Heading 2
STR
expected = [
[:HEADER, 1, 0, 0],
[:TEXT, 'Heading', 2, 0],
[:NEWLINE, "\n", 9, 0],
[:HEADER, 2, 0, 1],
[:TEXT, 'Heading 2', 3, 1],
[:NEWLINE, "\n", 12, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_heading_empty
str = <<-STR
===
* bullet
STR
expected = [
[:HEADER, 3, 0, 0],
[:NEWLINE, "\n", 3, 0],
[:BULLET, "*", 0, 1],
[:TEXT, "bullet", 2, 1],
[:NEWLINE, "\n", 8, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_heading_heading
str = <<-STR
= =
STR
expected = [
[:HEADER, 1, 0, 0],
[:TEXT, '=', 2, 0],
[:NEWLINE, "\n", 3, 0],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_heading_no_space
str = <<-STR
=Heading
==Heading 2
STR
expected = [
[:HEADER, 1, 0, 0],
[:TEXT, 'Heading', 1, 0],
[:NEWLINE, "\n", 8, 0],
[:HEADER, 2, 0, 1],
[:TEXT, 'Heading 2', 2, 1],
[:NEWLINE, "\n", 11, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_label
str = <<-STR
[cat] l1
[dog] l1.1
STR
expected = [
[:LABEL, 'cat', 0, 0],
[:TEXT, 'l1', 6, 0],
[:NEWLINE, "\n", 8, 0],
[:LABEL, 'dog', 0, 1],
[:TEXT, 'l1.1', 6, 1],
[:NEWLINE, "\n", 10, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_label_note
str = <<-STR
[label]
note::
STR
expected = [
[:LABEL, 'label', 0, 0],
[:NEWLINE, "\n", 7, 0],
[:NOTE, 'note', 2, 1],
[:NEWLINE, "\n", 8, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_label_newline
str = <<-STR
[cat]
l1
STR
expected = [
[:LABEL, 'cat', 0, 0],
[:NEWLINE, "\n", 5, 0],
[:TEXT, 'l1', 2, 1],
[:NEWLINE, "\n", 4, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_label_newline_windows
str = <<-STR
[cat]\r
l1\r
STR
expected = [
[:LABEL, 'cat', 0, 0],
[:NEWLINE, "\n", 6, 0],
[:TEXT, 'l1', 2, 1],
[:NEWLINE, "\n", 5, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_lalpha
str = <<-STR
a. l1
b. l1.1
STR
expected = [
[:LALPHA, 'a', 0, 0],
[:TEXT, 'l1', 3, 0],
[:NEWLINE, "\n", 5, 0],
[:LALPHA, 'b', 0, 1],
[:TEXT, 'l1.1', 3, 1],
[:NEWLINE, "\n", 7, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_line_break
str = "now is\nthe time \nfor all\n"
expected = [
[:TEXT, 'now is', 0, 0],
[:NEWLINE, "\n", 6, 0],
[:TEXT, 'the time', 0, 1],
[:BREAK, " ", 8, 1],
[:NEWLINE, "\n", 10, 1],
[:TEXT, 'for all', 0, 2],
[:NEWLINE, "\n", 7, 2],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_line_break_long
str = "now is\nthe time \nfor all\n"
expected = [
[:TEXT, 'now is', 0, 0],
[:NEWLINE, "\n", 6, 0],
[:TEXT, 'the time ', 0, 1],
[:BREAK, ' ', 9, 1],
[:NEWLINE, "\n", 11, 1],
[:TEXT, 'for all', 0, 2],
[:NEWLINE, "\n", 7, 2],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_line_break_no_short
str = "now is\nthe time \nfor all\n"
expected = [
[:TEXT, 'now is', 0, 0],
[:NEWLINE, "\n", 6, 0],
[:TEXT, 'the time ', 0, 1],
[:NEWLINE, "\n", 9, 1],
[:TEXT, 'for all', 0, 2],
[:NEWLINE, "\n", 7, 2],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_note
str = <<-STR
cat:: l1
dog:: l1.1
STR
expected = [
[:NOTE, 'cat', 0, 0],
[:TEXT, 'l1', 6, 0],
[:NEWLINE, "\n", 8, 0],
[:NOTE, 'dog', 0, 1],
[:TEXT, 'l1.1', 6, 1],
[:NEWLINE, "\n", 10, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_note_empty
str = <<-STR
cat::
dog::
STR
expected = [
[:NOTE, 'cat', 0, 0],
[:NEWLINE, "\n", 5, 0],
[:NOTE, 'dog', 0, 1],
[:NEWLINE, "\n", 5, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_note_newline
str = <<-STR
cat::
l1
STR
expected = [
[:NOTE, 'cat', 0, 0],
[:NEWLINE, "\n", 5, 0],
[:TEXT, 'l1', 2, 1],
[:NEWLINE, "\n", 4, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_note_utf_8
skip 'Encoding not implemented' unless Object.const_defined? :Encoding
str = <<-STR
cät:: l1a
l1b
døg:: l2a
l2b
STR
expected = [
[:NOTE, 'cät', 0, 0],
[:TEXT, 'l1a', 6, 0],
[:NEWLINE, "\n", 9, 0],
[:TEXT, 'l1b', 6, 1],
[:NEWLINE, "\n", 9, 1],
[:NOTE, 'døg', 0, 2],
[:TEXT, 'l2a', 6, 2],
[:NEWLINE, "\n", 9, 2],
[:TEXT, 'l2b', 6, 3],
[:NEWLINE, "\n", 9, 3],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_note_newline_windows
str = <<-STR
cat::\r
l1\r
STR
expected = [
[:NOTE, 'cat', 0, 0],
[:NEWLINE, "\n", 6, 0],
[:TEXT, 'l1', 2, 1],
[:NEWLINE, "\n", 5, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_note_not
str = <<-STR
Cat::Dog
STR
expected = [
[:TEXT, 'Cat::Dog', 0, 0],
[:NEWLINE, "\n", 8, 0],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_number
str = <<-STR
1. l1
2. l1.1
STR
expected = [
[:NUMBER, '1', 0, 0],
[:TEXT, 'l1', 3, 0],
[:NEWLINE, "\n", 5, 0],
[:NUMBER, '2', 0, 1],
[:TEXT, 'l1.1', 3, 1],
[:NEWLINE, "\n", 7, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_number_period
str = <<-STR
1. blah blah blah
l.
2. blah blah blah blah
d.
STR
expected = [
[:NUMBER, "1", 0, 0],
[:TEXT, "blah blah blah", 3, 0],
[:NEWLINE, "\n", 17, 0],
[:TEXT, "l.", 3, 1],
[:NEWLINE, "\n", 5, 1],
[:NUMBER, "2", 0, 2],
[:TEXT, "blah blah blah blah", 3, 2],
[:NEWLINE, "\n", 22, 2],
[:TEXT, "d.", 3, 3],
[:NEWLINE, "\n", 5, 3]
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_number_period_continue
str = <<-STR
1. blah blah blah
l. more stuff
2. blah blah blah blah
d. other stuff
STR
expected = [
[:NUMBER, "1", 0, 0],
[:TEXT, "blah blah blah", 3, 0],
[:NEWLINE, "\n", 17, 0],
[:LALPHA, "l", 3, 1],
[:TEXT, "more stuff", 7, 1],
[:NEWLINE, "\n", 17, 1],
[:NUMBER, "2", 0, 2],
[:TEXT, "blah blah blah blah", 3, 2],
[:NEWLINE, "\n", 22, 2],
[:LALPHA, "d", 3, 3],
[:TEXT, "other stuff", 6, 3],
[:NEWLINE, "\n", 17, 3]
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_paragraphs
str = <<-STR
now is
the time
for all
STR
expected = [
[:TEXT, 'now is', 0, 0],
[:NEWLINE, "\n", 6, 0],
[:TEXT, 'the time', 0, 1],
[:NEWLINE, "\n", 8, 1],
[:NEWLINE, "\n", 0, 2],
[:TEXT, 'for all', 0, 3],
[:NEWLINE, "\n", 7, 3],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_rule
str = <<-STR
---
--- blah ---
STR
expected = [
[:RULE, 1, 0, 0],
[:NEWLINE, "\n", 3, 0],
[:NEWLINE, "\n", 0, 1],
[:TEXT, "--- blah ---", 0, 2],
[:NEWLINE, "\n", 12, 2],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_rule_windows
str = <<-STR
---\r
--- blah ---\r
STR
expected = [
[:RULE, 1, 0, 0],
[:NEWLINE, "\n", 4, 0],
[:NEWLINE, "\n", 0, 1],
[:TEXT, "--- blah ---", 0, 2],
[:NEWLINE, "\n", 13, 2],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_ualpha
str = <<-STR
A. l1
B. l1.1
STR
expected = [
[:UALPHA, 'A', 0, 0],
[:TEXT, 'l1', 3, 0],
[:NEWLINE, "\n", 5, 0],
[:UALPHA, 'B', 0, 1],
[:TEXT, 'l1.1', 3, 1],
[:NEWLINE, "\n", 7, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_verbatim_heading
str = <<-STR
Example heading:
=== heading three
STR
expected = [
[:TEXT, 'Example heading:', 0, 0],
[:NEWLINE, "\n", 16, 0],
[:NEWLINE, "\n", 0, 1],
[:HEADER, 3, 3, 2],
[:TEXT, 'heading three', 7, 2],
[:NEWLINE, "\n", 20, 2],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_verbatim_rule
str = <<-STR
Verbatim section here that is double-underlined
===============================================
STR
expected = [
[:TEXT, 'Verbatim section here that is double-underlined', 2, 0],
[:NEWLINE, "\n", 49, 0],
[:HEADER, 47, 2, 1],
[:NEWLINE, "\n", 49, 1],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_tokenize_verbatim_rule_fancy
str = <<-STR
A
b
===============================================
c
STR
expected = [
[:TEXT, 'A', 2, 0],
[:NEWLINE, "\n", 3, 0],
[:TEXT, 'b', 4, 1],
[:NEWLINE, "\n", 5, 1],
[:HEADER, 47, 2, 2],
[:NEWLINE, "\n", 49, 2],
[:TEXT, 'c', 4, 3],
[:NEWLINE, "\n", 5, 3],
]
assert_equal expected, @RMP.tokenize(str)
end
def test_token_pos
parser = @RMP.new
s = parser.setup_scanner 'cät'
s.scan(/\S+/)
if Object.const_defined? :Encoding then
assert_equal [3, 0], parser.token_pos(s.pos)
else
assert_equal [4, 0], parser.token_pos(s.pos)
end
end
# HACK move to Verbatim test case
def test_verbatim_normalize
v = @RM::Verbatim.new "foo\n", "\n", "\n", "bar\n"
v.normalize
assert_equal ["foo\n", "\n", "bar\n"], v.parts
v = @RM::Verbatim.new "foo\n", "\n"
v.normalize
assert_equal ["foo\n"], v.parts
end
def test_unget
parser = util_parser
parser.get
parser.unget
assert_equal [:HEADER, 1, 0, 0], parser.peek_token
assert_raises @RMP::Error do
parser.unget
end
assert_equal 8, parser.tokens.length
end
def util_parser
str = <<-STR
= Heading
Some text here
some more text over here
STR
@parser = @RMP.new
@parser.tokenize str
@parser
end
end