bastodon/config/initializers/twitter_regex.rb
Claire b21f3aa21d
Minor memory optimizations (#16507)
Reduce constant memory usage by ~100kB and further reduce boot-up memory
allocations and temporary memory use by a further ~200kB.
2021-10-14 21:04:57 +02:00

104 lines
5.1 KiB
Ruby

module Twitter::TwitterText
class Configuration
def emoji_parsing_enabled
false
end
end
class Regex
REGEXEN[:valid_general_url_path_chars] = /[^\p{White_Space}<>\(\)\?]/iou
REGEXEN[:valid_url_path_ending_chars] = /[^\p{White_Space}\(\)\?!\*"'「」<>;:=\,\.\$%\[\]~&\|@]|(?:#{REGEXEN[:valid_url_balanced_parens]})/iou
REGEXEN[:valid_url_balanced_parens] = /
\(
(?:
#{REGEXEN[:valid_general_url_path_chars]}+
|
# allow one nested level of balanced parentheses
(?:
#{REGEXEN[:valid_general_url_path_chars]}*
\(
#{REGEXEN[:valid_general_url_path_chars]}+
\)
#{REGEXEN[:valid_general_url_path_chars]}*
)
)
\)
/iox
UCHARS = '\u{A0}-\u{D7FF}\u{F900}-\u{FDCF}\u{FDF0}-\u{FFEF}\u{10000}-\u{1FFFD}\u{20000}-\u{2FFFD}\u{30000}-\u{3FFFD}\u{40000}-\u{4FFFD}\u{50000}-\u{5FFFD}\u{60000}-\u{6FFFD}\u{70000}-\u{7FFFD}\u{80000}-\u{8FFFD}\u{90000}-\u{9FFFD}\u{A0000}-\u{AFFFD}\u{B0000}-\u{BFFFD}\u{C0000}-\u{CFFFD}\u{D0000}-\u{DFFFD}\u{E1000}-\u{EFFFD}\u{E000}-\u{F8FF}\u{F0000}-\u{FFFFD}\u{100000}-\u{10FFFD}'
REGEXEN[:valid_url_query_chars] = /[a-z0-9!?\*'\(\);:&=\+\$\/%#\[\]\-_\.,~|@#{UCHARS}]/iou
REGEXEN[:valid_url_query_ending_chars] = /[a-z0-9_&=#\/\-#{UCHARS}]/iou
REGEXEN[:valid_url_path] = /(?:
(?:
#{REGEXEN[:valid_general_url_path_chars]}*
(?:#{REGEXEN[:valid_url_balanced_parens]} #{REGEXEN[:valid_general_url_path_chars]}*)*
#{REGEXEN[:valid_url_path_ending_chars]}
)|(?:#{REGEXEN[:valid_general_url_path_chars]}+\/)
)/iox
REGEXEN[:valid_url] = %r{
( # $1 total match
(#{REGEXEN[:valid_url_preceding_chars]}) # $2 Preceding character
( # $3 URL
((?:https?|dat|dweb|ipfs|ipns|ssb|gopher|gemini):\/\/)? # $4 Protocol (optional)
(#{REGEXEN[:valid_domain]}) # $5 Domain(s)
(?::(#{REGEXEN[:valid_port_number]}))? # $6 Port number (optional)
(/#{REGEXEN[:valid_url_path]}*)? # $7 URL Path and anchor
(\?#{REGEXEN[:valid_url_query_chars]}*#{REGEXEN[:valid_url_query_ending_chars]})? # $8 Query String
)
)
}iox
REGEXEN[:validate_nodeid] = /(?:
#{REGEXEN[:validate_url_unreserved]}|
#{REGEXEN[:validate_url_pct_encoded]}|
[!$()*+,;=]
)/iox
REGEXEN[:validate_resid] = /(?:
#{REGEXEN[:validate_url_unreserved]}|
#{REGEXEN[:validate_url_pct_encoded]}|
#{REGEXEN[:validate_url_sub_delims]}
)/iox
REGEXEN[:valid_extended_uri] = %r{
( # $1 total match
(#{REGEXEN[:valid_url_preceding_chars]}) # $2 Preceding character
( # $3 URL
(
(xmpp:) # Protocol
(//#{REGEXEN[:validate_nodeid]}+@#{REGEXEN[:valid_domain]}/)? # Authority (optional)
(#{REGEXEN[:validate_nodeid]}+@)? # Username in path (optional)
(#{REGEXEN[:valid_domain]}) # Domain in path
(/#{REGEXEN[:validate_resid]}+)? # Resource in path (optional)
(\?#{REGEXEN[:valid_url_query_chars]}*#{REGEXEN[:valid_url_query_ending_chars]})? # Query String
) | (
(magnet:) # Protocol
(\?#{REGEXEN[:valid_url_query_chars]}*#{REGEXEN[:valid_url_query_ending_chars]}) # Query String
)
)
)
}iox
end
module Extractor
# Extracts a list of all XMPP and magnet URIs included in the Toot <tt>text</tt> along
# with the indices. If the <tt>text</tt> is <tt>nil</tt> or contains no
# XMPP or magnet URIs an empty array will be returned.
#
# If a block is given then it will be called for each XMPP URI.
def extract_extra_uris_with_indices(text, _options = {}) # :yields: uri, start, end
return [] unless text && text.index(":")
urls = []
text.to_s.scan(Twitter::TwitterText::Regex[:valid_extended_uri]) do
valid_uri_match_data = $~
start_position = valid_uri_match_data.char_begin(3)
end_position = valid_uri_match_data.char_end(3)
urls << {
:url => valid_uri_match_data[3],
:indices => [start_position, end_position]
}
end
urls.each{|url| yield url[:url], url[:indices].first, url[:indices].last} if block_given?
urls
end
end
end