270 lines
7.6 KiB
Ruby
270 lines
7.6 KiB
Ruby
require_dependency "onpdiff"
|
|
|
|
class DiscourseDiff
|
|
|
|
MAX_DIFFERENCE = 200
|
|
|
|
def initialize(before, after)
|
|
@before = before
|
|
@after = after
|
|
before_html = tokenize_html_blocks(@before)
|
|
after_html = tokenize_html_blocks(@after)
|
|
before_markdown = tokenize_line(CGI::escapeHTML(@before))
|
|
after_markdown = tokenize_line(CGI::escapeHTML(@after))
|
|
|
|
@block_by_block_diff = ONPDiff.new(before_html, after_html).diff
|
|
@line_by_line_diff = ONPDiff.new(before_markdown, after_markdown).short_diff
|
|
end
|
|
|
|
def inline_html
|
|
i = 0
|
|
inline = []
|
|
while i < @block_by_block_diff.length
|
|
op_code = @block_by_block_diff[i][1]
|
|
if op_code == :common then inline << @block_by_block_diff[i][0]
|
|
else
|
|
if op_code == :delete
|
|
opposite_op_code = :add
|
|
klass = "del"
|
|
first = i
|
|
second = i + 1
|
|
else
|
|
opposite_op_code = :delete
|
|
klass = "ins"
|
|
first = i + 1
|
|
second = i
|
|
end
|
|
|
|
if i + 1 < @block_by_block_diff.length && @block_by_block_diff[i + 1][1] == opposite_op_code
|
|
diff = ONPDiff.new(tokenize_html(@block_by_block_diff[first][0]), tokenize_html(@block_by_block_diff[second][0])).diff
|
|
inline << generate_inline_html(diff)
|
|
i += 1
|
|
else
|
|
inline << add_class_or_wrap_in_tags(@block_by_block_diff[i][0], klass)
|
|
end
|
|
end
|
|
i += 1
|
|
end
|
|
|
|
"<div class=\"inline-diff\">#{inline.join}</div>"
|
|
end
|
|
|
|
def side_by_side_html
|
|
i = 0
|
|
left, right = [], []
|
|
while i < @block_by_block_diff.length
|
|
op_code = @block_by_block_diff[i][1]
|
|
if op_code == :common
|
|
left << @block_by_block_diff[i][0]
|
|
right << @block_by_block_diff[i][0]
|
|
else
|
|
if op_code == :delete
|
|
opposite_op_code = :add
|
|
side = left
|
|
klass = "del"
|
|
first = i
|
|
second = i + 1
|
|
else
|
|
opposite_op_code = :delete
|
|
side = right
|
|
klass = "ins"
|
|
first = i + 1
|
|
second = i
|
|
end
|
|
|
|
if i + 1 < @block_by_block_diff.length && @block_by_block_diff[i + 1][1] == opposite_op_code
|
|
diff = ONPDiff.new(tokenize_html(@block_by_block_diff[first][0]), tokenize_html(@block_by_block_diff[second][0])).diff
|
|
deleted, inserted = generate_side_by_side_html(diff)
|
|
left << deleted
|
|
right << inserted
|
|
i += 1
|
|
else
|
|
side << add_class_or_wrap_in_tags(@block_by_block_diff[i][0], klass)
|
|
end
|
|
end
|
|
i += 1
|
|
end
|
|
|
|
"<div class=\"revision-content\">#{left.join}</div><div class=\"revision-content\">#{right.join}</div>"
|
|
end
|
|
|
|
def side_by_side_markdown
|
|
i = 0
|
|
table = ["<table class=\"markdown\">"]
|
|
while i < @line_by_line_diff.length
|
|
table << "<tr>"
|
|
op_code = @line_by_line_diff[i][1]
|
|
if op_code == :common
|
|
table << "<td>#{@line_by_line_diff[i][0]}</td>"
|
|
table << "<td>#{@line_by_line_diff[i][0]}</td>"
|
|
else
|
|
if op_code == :delete
|
|
opposite_op_code = :add
|
|
first = i
|
|
second = i + 1
|
|
else
|
|
opposite_op_code = :delete
|
|
first = i + 1
|
|
second = i
|
|
end
|
|
|
|
if i + 1 < @line_by_line_diff.length && @line_by_line_diff[i + 1][1] == opposite_op_code
|
|
before_tokens, after_tokens = tokenize_markdown(@line_by_line_diff[first][0]), tokenize_markdown(@line_by_line_diff[second][0])
|
|
if (before_tokens.length - after_tokens.length).abs > MAX_DIFFERENCE
|
|
before_tokens, after_tokens = tokenize_line(@line_by_line_diff[first][0]), tokenize_line(@line_by_line_diff[second][0])
|
|
end
|
|
diff = ONPDiff.new(before_tokens, after_tokens).short_diff
|
|
deleted, inserted = generate_side_by_side_markdown(diff)
|
|
table << "<td class=\"diff-del\">#{deleted.join}</td>"
|
|
table << "<td class=\"diff-ins\">#{inserted.join}</td>"
|
|
i += 1
|
|
else
|
|
if op_code == :delete
|
|
table << "<td class=\"diff-del\">#{@line_by_line_diff[i][0]}</td>"
|
|
table << "<td></td>"
|
|
else
|
|
table << "<td></td>"
|
|
table << "<td class=\"diff-ins\">#{@line_by_line_diff[i][0]}</td>"
|
|
end
|
|
end
|
|
end
|
|
table << "</tr>"
|
|
i += 1
|
|
end
|
|
table << "</table>"
|
|
|
|
table.join
|
|
end
|
|
|
|
private
|
|
|
|
def tokenize_line(text)
|
|
text.scan(/[^\r\n]+[\r\n]*/)
|
|
end
|
|
|
|
def tokenize_markdown(text)
|
|
t, tokens = [], []
|
|
i = 0
|
|
while i < text.length
|
|
if text[i] =~ /\w/
|
|
t << text[i]
|
|
elsif text[i] =~ /[ \t]/ && t.join =~ /^\w+$/
|
|
begin
|
|
t << text[i]
|
|
i += 1
|
|
end while i < text.length && text[i] =~ /[ \t]/
|
|
i -= 1
|
|
tokens << t.join
|
|
t = []
|
|
else
|
|
tokens << t.join if t.length > 0
|
|
tokens << text[i]
|
|
t = []
|
|
end
|
|
i += 1
|
|
end
|
|
tokens << t.join if t.length > 0
|
|
tokens
|
|
end
|
|
|
|
def tokenize_html_blocks(html)
|
|
Nokogiri::HTML.fragment(html).search("./*").map(&:to_html)
|
|
end
|
|
|
|
def tokenize_html(html)
|
|
HtmlTokenizer.tokenize(html)
|
|
end
|
|
|
|
def add_class_or_wrap_in_tags(html_or_text, klass)
|
|
index_of_next_chevron = html_or_text.index(">")
|
|
if html_or_text.length > 0 && html_or_text[0] == '<' && index_of_next_chevron
|
|
index_of_class = html_or_text.index("class=")
|
|
if index_of_class.nil? || index_of_class > index_of_next_chevron
|
|
# we do not have a class for the current tag
|
|
# add it right before the ">"
|
|
html_or_text.insert(index_of_next_chevron, " class=\"diff-#{klass}\"")
|
|
else
|
|
# we have a class, insert it at the beginning
|
|
html_or_text.insert(index_of_class + "class=".length + 1, "diff-#{klass} ")
|
|
end
|
|
else
|
|
"<#{klass}>#{html_or_text}</#{klass}>"
|
|
end
|
|
end
|
|
|
|
def generate_inline_html(diff)
|
|
inline = []
|
|
diff.each do |d|
|
|
case d[1]
|
|
when :common then inline << d[0]
|
|
when :delete then inline << add_class_or_wrap_in_tags(d[0], "del")
|
|
when :add then inline << add_class_or_wrap_in_tags(d[0], "ins")
|
|
end
|
|
end
|
|
inline
|
|
end
|
|
|
|
def generate_side_by_side_html(diff)
|
|
deleted, inserted = [], []
|
|
diff.each do |d|
|
|
case d[1]
|
|
when :common
|
|
deleted << d[0]
|
|
inserted << d[0]
|
|
when :delete then deleted << add_class_or_wrap_in_tags(d[0], "del")
|
|
when :add then inserted << add_class_or_wrap_in_tags(d[0], "ins")
|
|
end
|
|
end
|
|
[deleted, inserted]
|
|
end
|
|
|
|
def generate_side_by_side_markdown(diff)
|
|
deleted, inserted = [], []
|
|
diff.each do |d|
|
|
case d[1]
|
|
when :common
|
|
deleted << d[0]
|
|
inserted << d[0]
|
|
when :delete then deleted << "<del>#{d[0]}</del>"
|
|
when :add then inserted << "<ins>#{d[0]}</ins>"
|
|
end
|
|
end
|
|
[deleted, inserted]
|
|
end
|
|
|
|
class HtmlTokenizer < Nokogiri::XML::SAX::Document
|
|
|
|
attr_accessor :tokens
|
|
|
|
def initialize
|
|
@tokens = []
|
|
end
|
|
|
|
def self.tokenize(html)
|
|
me = new
|
|
parser = Nokogiri::HTML::SAX::Parser.new(me)
|
|
parser.parse("<html><body>#{html}</body></html>")
|
|
me.tokens
|
|
end
|
|
|
|
USELESS_TAGS = %w{html body}
|
|
def start_element(name, attributes = [])
|
|
return if USELESS_TAGS.include?(name)
|
|
attrs = attributes.map { |a| " #{a[0]}=\"#{a[1]}\"" }.join
|
|
@tokens << "<#{name}#{attrs}>"
|
|
end
|
|
|
|
AUTOCLOSING_TAGS = %w{area base br col embed hr img input meta}
|
|
def end_element(name)
|
|
return if USELESS_TAGS.include?(name) || AUTOCLOSING_TAGS.include?(name)
|
|
@tokens << "</#{name}>"
|
|
end
|
|
|
|
def characters(string)
|
|
@tokens.concat string.scan(/\W|\w+[ \t]*/).map { |x| CGI::escapeHTML(x) }
|
|
end
|
|
|
|
end
|
|
|
|
end
|