2019-05-03 06:17:27 +08:00
|
|
|
|
# frozen_string_literal: true
|
|
|
|
|
|
2017-12-06 08:47:31 +08:00
|
|
|
|
class PlainTextToMarkdown
|
2024-11-06 06:27:49 +08:00
|
|
|
|
SIGNATURE_SEPARATOR = "-- "
|
2017-12-06 08:47:31 +08:00
|
|
|
|
|
|
|
|
|
def initialize(plaintext, opts = {})
|
|
|
|
|
@plaintext = plaintext
|
|
|
|
|
@lines = []
|
|
|
|
|
|
|
|
|
|
@format_flowed = opts[:format_flowed] || false
|
|
|
|
|
@delete_flowed_space = opts[:delete_flowed_space] || false
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def to_markdown
|
|
|
|
|
prepare_lines
|
|
|
|
|
classify_lines
|
|
|
|
|
|
2019-05-03 06:17:27 +08:00
|
|
|
|
markdown = +""
|
2017-12-06 08:47:31 +08:00
|
|
|
|
last_quote_level = 0
|
|
|
|
|
last_line_blank = false
|
|
|
|
|
|
|
|
|
|
@lines.each do |line|
|
|
|
|
|
current_line_blank = line.text.blank?
|
|
|
|
|
|
|
|
|
|
unless last_line_blank && current_line_blank
|
|
|
|
|
if line.quote_level > 0
|
|
|
|
|
quote_identifiers = ">" * line.quote_level
|
|
|
|
|
unless line.quote_level >= last_quote_level || current_line_blank
|
|
|
|
|
markdown << quote_identifiers << "\n"
|
2023-01-09 20:10:19 +08:00
|
|
|
|
end
|
2017-12-06 08:47:31 +08:00
|
|
|
|
markdown << quote_identifiers
|
|
|
|
|
markdown << " " unless current_line_blank
|
|
|
|
|
else
|
|
|
|
|
markdown << "\n" unless last_quote_level == 0 || current_line_blank
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
markdown << convert_text(line)
|
|
|
|
|
markdown << "\n"
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
last_line_blank = current_line_blank
|
|
|
|
|
last_quote_level = line.quote_level
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
markdown.rstrip!
|
|
|
|
|
markdown
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
private
|
|
|
|
|
|
|
|
|
|
class CodeBlock < Struct.new(:start_line, :end_line)
|
|
|
|
|
def initialize(start_line, end_line = nil)
|
|
|
|
|
super
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def valid?
|
|
|
|
|
start_line.present? && end_line.present?
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
class Line < Struct.new(:text, :quote_level, :code_block)
|
|
|
|
|
def initialize(text, quote_level = 0, code_block = nil)
|
|
|
|
|
super
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def valid_code_block?
|
|
|
|
|
code_block&.valid?
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def prepare_lines
|
|
|
|
|
previous_line = nil
|
|
|
|
|
|
|
|
|
|
@plaintext.each_line do |text|
|
|
|
|
|
text.chomp!
|
|
|
|
|
line = Line.new(text)
|
|
|
|
|
|
|
|
|
|
remove_quote_level_indicators!(line)
|
|
|
|
|
|
|
|
|
|
if @format_flowed
|
|
|
|
|
line = merge_lines(line, previous_line)
|
|
|
|
|
@lines << line unless line == previous_line
|
|
|
|
|
else
|
|
|
|
|
@lines << line
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
previous_line = line
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def classify_lines
|
|
|
|
|
previous_line = nil
|
|
|
|
|
|
|
|
|
|
@lines.each do |line|
|
|
|
|
|
classify_line_as_code!(line, previous_line)
|
|
|
|
|
|
|
|
|
|
previous_line = line
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
# @param line [Line]
|
|
|
|
|
def remove_quote_level_indicators!(line)
|
2023-01-21 02:52:49 +08:00
|
|
|
|
match_data = line.text.match(/\A(?<indicators>>+)\s?(?<text>.*)/)
|
2017-12-06 08:47:31 +08:00
|
|
|
|
|
|
|
|
|
if match_data
|
|
|
|
|
line.text = match_data[:text]
|
|
|
|
|
line.quote_level = match_data[:indicators].length
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
# @param line [Line]
|
|
|
|
|
# @param previous_line [Line]
|
|
|
|
|
# @return [Line]
|
|
|
|
|
def merge_lines(line, previous_line)
|
|
|
|
|
return line if previous_line.nil? || line.text.blank?
|
|
|
|
|
return line if line.text == SIGNATURE_SEPARATOR || previous_line.text == SIGNATURE_SEPARATOR
|
|
|
|
|
unless line.quote_level == previous_line.quote_level && previous_line.text.end_with?(" ")
|
|
|
|
|
return line
|
2023-01-09 20:10:19 +08:00
|
|
|
|
end
|
2017-12-06 08:47:31 +08:00
|
|
|
|
|
|
|
|
|
previous_line.text = previous_line.text[0...-1] if @delete_flowed_space
|
|
|
|
|
previous_line.text += line.text
|
|
|
|
|
previous_line
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
# @param line [Line]
|
|
|
|
|
# @param previous_line [Line]
|
|
|
|
|
def classify_line_as_code!(line, previous_line)
|
|
|
|
|
line.code_block = previous_line.code_block unless previous_line.nil? ||
|
|
|
|
|
previous_line.valid_code_block?
|
2023-01-21 02:52:49 +08:00
|
|
|
|
return unless line.text =~ /\A\s{0,3}```/
|
2017-12-06 08:47:31 +08:00
|
|
|
|
|
|
|
|
|
if line.code_block.present?
|
|
|
|
|
line.code_block.end_line = line
|
|
|
|
|
else
|
|
|
|
|
line.code_block = CodeBlock.new(line)
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
# @param line [Line]
|
|
|
|
|
# @return [string]
|
|
|
|
|
def convert_text(line)
|
|
|
|
|
text = line.text
|
|
|
|
|
|
|
|
|
|
if line.valid_code_block?
|
|
|
|
|
code_block = line.code_block
|
|
|
|
|
return code_block.start_line == line || code_block.end_line == line ? text.lstrip : text
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
converted_text = replace_duplicate_links(text)
|
|
|
|
|
converted_text = escape_special_characters(converted_text)
|
|
|
|
|
converted_text = indent_with_non_breaking_spaces(converted_text)
|
|
|
|
|
converted_text
|
|
|
|
|
end
|
|
|
|
|
|
2024-10-16 10:09:07 +08:00
|
|
|
|
URL_REGEX = URI.regexp(%w[http https ftp mailto])
|
|
|
|
|
BEFORE = Regexp.escape(%Q|([<«"“'‘|)
|
|
|
|
|
AFTER = Regexp.escape(%Q|)]>»"”'’|)
|
2019-06-07 07:26:06 +08:00
|
|
|
|
|
2017-12-06 08:47:31 +08:00
|
|
|
|
def replace_duplicate_links(text)
|
2019-06-07 07:26:06 +08:00
|
|
|
|
urls = Set.new
|
|
|
|
|
text.scan(URL_REGEX) { urls << $& }
|
2017-12-06 08:47:31 +08:00
|
|
|
|
|
2019-06-07 07:26:06 +08:00
|
|
|
|
urls.each do |url|
|
2019-06-04 14:20:40 +08:00
|
|
|
|
escaped = Regexp.escape(url)
|
2019-06-07 07:26:06 +08:00
|
|
|
|
text.gsub!(
|
|
|
|
|
Regexp.new(%Q|#{escaped}\s*[#{BEFORE}]?#{escaped}[#{AFTER}]?|, Regexp::IGNORECASE),
|
|
|
|
|
url,
|
|
|
|
|
)
|
2017-12-06 08:47:31 +08:00
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
text
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def indent_with_non_breaking_spaces(text)
|
2023-01-21 02:52:49 +08:00
|
|
|
|
text.sub(/\A\s+/) do |s|
|
2017-12-06 08:47:31 +08:00
|
|
|
|
# replace tabs with 2 spaces
|
|
|
|
|
s.gsub!("\t", " ")
|
|
|
|
|
|
|
|
|
|
# replace indentation with non-breaking spaces
|
|
|
|
|
s.length > 1 ? " " * s.length : s
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def escape_special_characters(text)
|
2019-06-07 07:26:06 +08:00
|
|
|
|
urls = Set.new
|
|
|
|
|
text.scan(URL_REGEX) { urls << $& }
|
2017-12-06 08:47:31 +08:00
|
|
|
|
|
2019-06-07 07:26:06 +08:00
|
|
|
|
hoisted = urls.map { |url| [SecureRandom.hex, url] }.to_h
|
2017-12-06 08:47:31 +08:00
|
|
|
|
|
2019-06-07 07:26:06 +08:00
|
|
|
|
hoisted.each { |h, url| text.gsub!(url, h) }
|
|
|
|
|
|
|
|
|
|
text.gsub!(/[\\`*_{}\[\]()#+\-.!~]/) { |c| "\\#{c}" }
|
|
|
|
|
text = CGI.escapeHTML(text)
|
|
|
|
|
|
|
|
|
|
hoisted.each { |h, url| text.gsub!(h, url) }
|
|
|
|
|
|
|
|
|
|
text
|
2017-12-06 08:47:31 +08:00
|
|
|
|
end
|
|
|
|
|
end
|