Performance issue in regex for XML stylesheet parsing

[connorshea]

In the RSS xml stylesheet code there is a regex with backtracking that results in an O(N^2) computation when parsing specifically-shaped RSS feeds.

Specifically, the CONTENT_PATTERN regex at lib/rss/parser.rb:394.

([^=]+)= in that regex greedily consumes characters and backtracks quadratically.

Example code:

require "rss"
require "benchmark"

[10000, 20000, 40000, 80000].each do |n|
 payload = "a" * n
 feed = %Q{<?xml-stylesheet #{payload}=?>\n<rss version="2.0"><channel><title>t</title><link>http://x</link><description>d</description></channel></rss>}
 t = Benchmark.realtime { RSS::Parser.parse(feed, false) rescue nil }
 printf("n=%d total_parse=%.4fs\n", n, t)
end

Output on Ruby 3.4.9:

n=10000 total_parse=0.2715s
n=20000 total_parse=1.0939s
n=40000 total_parse=4.3082s
n=80000 total_parse=17.7647s

You can easily end up with multi-minute parsing times on specific XML, so we should probably fix that to improve the parse performance.