ruby/test/rdoc/test_rdoc_text.rb

542 строки
11 KiB
Ruby
Исходник Ответственный История

Этот файл содержит неоднозначные символы Юникода!

Этот файл содержит неоднозначные символы Юникода, которые могут быть перепутаны с другими в текущей локали. Если это намеренно, можете спокойно проигнорировать это предупреждение. Используйте кнопку Экранировать, чтобы подсветить эти символы.

# coding: utf-8
# frozen_string_literal: false
require 'rdoc/test_case'
class TestRDocText < RDoc::TestCase
include RDoc::Text
def setup
super
@options = RDoc::Options.new
@top_level = @store.add_file 'file.rb'
end
def test_self_encode_fallback
assert_equal '…',
RDoc::Text::encode_fallback('…', Encoding::UTF_8, '...')
assert_equal '...',
RDoc::Text::encode_fallback('…', Encoding::US_ASCII, '...')
end
def test_expand_tabs
assert_equal("hello\n dave",
expand_tabs("hello\n dave"), 'spaces')
assert_equal("hello\n dave",
expand_tabs("hello\n\tdave"), 'tab')
assert_equal("hello\n dave",
expand_tabs("hello\n \tdave"), '1 space tab')
assert_equal("hello\n dave",
expand_tabs("hello\n \tdave"), '2 space tab')
assert_equal("hello\n dave",
expand_tabs("hello\n \tdave"), '3 space tab')
assert_equal("hello\n dave",
expand_tabs("hello\n \tdave"), '4 space tab')
assert_equal("hello\n dave",
expand_tabs("hello\n \tdave"), '5 space tab')
assert_equal("hello\n dave",
expand_tabs("hello\n \tdave"), '6 space tab')
assert_equal("hello\n dave",
expand_tabs("hello\n \tdave"), '7 space tab')
assert_equal("hello\n dave",
expand_tabs("hello\n \tdave"), '8 space tab')
assert_equal('. .',
expand_tabs(".\t\t."), 'dot tab tab dot')
assert_equal('a a',
Timeout.timeout(1) {expand_tabs("\ra\ta")}, "carriage return")
end
def test_expand_tabs_encoding
inn = "hello\ns\tdave"
inn.force_encoding Encoding::BINARY
out = expand_tabs inn
assert_equal "hello\ns dave", out
assert_equal Encoding::BINARY, out.encoding
end
def test_flush_left
text = <<-TEXT
we don't worry too much.
The comments associated with
TEXT
expected = <<-EXPECTED
we don't worry too much.
The comments associated with
EXPECTED
assert_equal expected, flush_left(text)
end
def test_flush_left_encoding
text = <<-TEXT
we don't worry too much.
The comments associated with
TEXT
text.force_encoding Encoding::US_ASCII
expected = <<-EXPECTED
we don't worry too much.
The comments associated with
EXPECTED
result = flush_left text
assert_equal expected, result
assert_equal Encoding::US_ASCII, result.encoding
end
def test_markup_string
out = markup('hi').gsub("\n", '')
assert_equal '<p>hi</p>', out
end
def test_markup_comment
out = markup(comment('hi')).gsub("\n", '')
assert_equal '<p>hi</p>', out
end
def test_normalize_comment_hash
text = <<-TEXT
##
# we don't worry too much.
#
# The comments associated with
TEXT
expected = <<-EXPECTED.rstrip
we don't worry too much.
The comments associated with
EXPECTED
assert_equal expected, normalize_comment(text)
end
def test_normalize_comment_stars_single_space
text = <<-TEXT
/*
* we don't worry too much.
*
* The comments associated with
*/
TEXT
expected = <<-EXPECTED.rstrip
we don't worry too much.
The comments associated with
EXPECTED
assert_equal expected, normalize_comment(text)
end
def test_normalize_comment_stars_single_double_space
text = <<-TEXT
/*
* we don't worry too much.
*
* The comments associated with
*/
TEXT
expected = <<-EXPECTED.rstrip
we don't worry too much.
The comments associated with
EXPECTED
assert_equal expected, normalize_comment(text)
end
def test_parse
assert_kind_of RDoc::Markup::Document, parse('hi')
end
def test_parse_comment
expected = RDoc::Markup::Document.new
expected.file = @top_level
c = comment ''
parsed = parse c
assert_equal expected, parsed
assert_same parsed, parse(c)
end
def test_parse_document
assert_equal RDoc::Markup::Document.new, parse(RDoc::Markup::Document.new)
end
def test_parse_empty
assert_equal RDoc::Markup::Document.new, parse('')
end
def test_parse_empty_newline
assert_equal RDoc::Markup::Document.new, parse("#\n")
end
def test_parse_format_markdown
expected =
@RM::Document.new(
@RM::Paragraph.new('it _works_'))
parsed = parse 'it *works*', 'markdown'
assert_equal expected, parsed
end
def test_parse_format_rd
expected =
@RM::Document.new(
@RM::Paragraph.new('it <em>works</em>'))
parsed = parse 'it ((*works*))', 'rd'
assert_equal expected, parsed
end
def test_parse_format_tomdoc
code = verb('1 + 1')
code.format = :ruby
expected =
doc(
para('It does a thing'),
blank_line,
head(3, 'Examples'),
blank_line,
code)
text = <<-TOMDOC
It does a thing
Examples
1 + 1
TOMDOC
parsed = parse text, 'tomdoc'
assert_equal expected, parsed
end
def test_parse_newline
assert_equal RDoc::Markup::Document.new, parse("\n")
end
def test_snippet
text = <<-TEXT
This is one-hundred characters or more of text in a single paragraph. This
paragraph will be cut off some point after the one-hundredth character.
TEXT
expected = <<-EXPECTED
<p>This is one-hundred characters or more of text in a single paragraph. This
paragraph will be cut off …
EXPECTED
assert_equal expected, snippet(text)
end
def test_snippet_comment
c = comment 'This is a comment'
assert_equal "<p>This is a comment\n", snippet(c)
end
def test_snippet_short
text = 'This is a comment'
assert_equal "<p>#{text}\n", snippet(text)
end
def test_strip_hashes
text = <<-TEXT
##
# we don't worry too much.
#
# The comments associated with
TEXT
expected = <<-EXPECTED
we don't worry too much.
The comments associated with
EXPECTED
assert_equal expected, strip_hashes(text)
end
def test_strip_hashes_encoding
text = <<-TEXT
##
# we don't worry too much.
#
# The comments associated with
TEXT
text.force_encoding Encoding::CP852
expected = <<-EXPECTED
we don't worry too much.
The comments associated with
EXPECTED
stripped = strip_hashes text
assert_equal expected, stripped
assert_equal Encoding::CP852, stripped.encoding
end
def test_strip_newlines
assert_equal ' ', strip_newlines("\n \n")
assert_equal 'hi', strip_newlines("\n\nhi")
assert_equal 'hi', strip_newlines( "hi\n\n")
assert_equal 'hi', strip_newlines("\n\nhi\n\n")
end
def test_strip_newlines_encoding
assert_equal Encoding::UTF_8, ''.encoding, 'Encoding sanity check'
text = " \n"
text.force_encoding Encoding::US_ASCII
stripped = strip_newlines text
assert_equal ' ', stripped
assert_equal Encoding::US_ASCII, stripped.encoding
end
def test_strip_stars
text = <<-TEXT
/*
* * we don't worry too much.
*
* The comments associated with
*/
TEXT
expected = <<-EXPECTED
* we don't worry too much.
The comments associated with
EXPECTED
assert_equal expected, strip_stars(text)
end
def test_strip_stars_document_method
text = <<-TEXT
/*
* Document-method: Zlib::GzipFile#mtime=
*
* A comment
*/
TEXT
expected = <<-EXPECTED
A comment
EXPECTED
assert_equal expected, strip_stars(text)
end
def test_strip_stars_encoding
text = <<-TEXT
/*
* * we don't worry too much.
*
* The comments associated with
*/
TEXT
text.force_encoding Encoding::CP852
expected = <<-EXPECTED
* we don't worry too much.
The comments associated with
EXPECTED
result = strip_stars text
assert_equal expected, result
assert_equal Encoding::CP852, result.encoding
end
def test_strip_stars_encoding2
text = <<-TEXT
/*
* * we don't worry too much.
*
* The comments associated with
*/
TEXT
text.force_encoding Encoding::BINARY
expected = <<-EXPECTED
* we don't worry too much.
The comments associated with
EXPECTED
result = strip_stars text
assert_equal expected, result
assert_equal Encoding::BINARY, result.encoding
end
def test_strip_stars_no_stars
text = <<-TEXT
* we don't worry too much.
The comments associated with
TEXT
expected = <<-EXPECTED
* we don't worry too much.
The comments associated with
EXPECTED
assert_equal expected, strip_stars(text)
end
def test_to_html_apostrophe
assert_equal 'a', to_html("'a")
assert_equal 'a', to_html("a'")
assert_equal 'a ', to_html("'a' '")
end
def test_to_html_backslash
assert_equal 'S', to_html('\\S')
end
def test_to_html_br
assert_equal '<br>', to_html('<br>')
end
def test_to_html_copyright
assert_equal '©', to_html('(c)')
end
def test_to_html_dash
assert_equal '-', to_html('-')
assert_equal '–', to_html('--')
assert_equal '—', to_html('---')
assert_equal '—-', to_html('----')
end
def test_to_html_double_backtick
assert_equal '“a', to_html('``a')
assert_equal '“a“', to_html('``a``')
end
def test_to_html_double_quote
assert_equal '“a', to_html('"a')
assert_equal 'a', to_html('"a"')
end
def test_to_html_double_quote_quot
assert_equal 'a', to_html('&quot;a')
assert_equal 'a', to_html('&quot;a&quot;')
end
def test_to_html_double_tick
assert_equal 'a', to_html("''a")
assert_equal 'a', to_html("''a''")
end
def test_to_html_ellipsis
assert_equal '..', to_html('..')
assert_equal '', to_html('...')
assert_equal '.', to_html('....')
end
def test_to_html_encoding
s = '...(c)'.encode Encoding::Shift_JIS
html = to_html s
assert_equal Encoding::Shift_JIS, html.encoding
expected = '(c)'.encode Encoding::Shift_JIS
assert_equal expected, html
end
def test_to_html_html_tag
assert_equal '<a href="http://example">his</a>',
to_html('<a href="http://example">hi\'s</a>')
end
def test_to_html_registered_trademark
assert_equal '®', to_html('(r)')
end
def test_to_html_tt_tag
assert_equal '<tt>hi\'s</tt>', to_html('<tt>hi\'s</tt>')
assert_equal '<tt>hi\\\'s</tt>', to_html('<tt>hi\\\\\'s</tt>')
end
def test_to_html_tt_tag_mismatch
_, err = verbose_capture_io do
assert_equal '<tt>hi', to_html('<tt>hi')
end
assert_equal "mismatched <tt> tag\n", err
end
def formatter
RDoc::Markup::ToHtml.new @options
end
def options
@options
end
end