Feature #19102
closedOptimize ERB::Util.html_escape more than CGI.escapeHTML for template engines
Description
Proposal¶
Change the behavior of ERB::Util.html_escape
in the following two parts:
- Skip converting an argument with
#to_s
if the argument is already aT_STRING
. - Do not allocate and return a new String when nothing needs to be escaped.
Background¶
The current ERB::Util.html_escape
is implemented as CGI.escapeHTML(s.to_s)
. So the performance is almost equal to CGI.escapeHTML
except for the to_s
call. Because it's common to embed non-String expressions in template engines, a template engine typically calls to_s
to convert non-String expressions to a String before escaping it, which is why the difference exists. Proposal (1) is useful for optimizing the case that something that's already a String
is embedded. We ignore the extreme case that String#to_s
is weirdly monkey-patched.
As to proposal (2), my original implementation of CGI.escapeHTML
https://github.com/ruby/ruby/pull/1164 was not calling rb_str_dup
for that case. However, because [Bug #11858] claimed returning the argument object for non-escaped cases is a backward incompatibility with the old gsub
-based implementation, we added the unneeded rb_str_dup
call and the performance for that case has been compromised. This behavior is completely unnecessary for template engines. On the other hand, because ERB::Util.html_escape
is a helper for ERB, we should not need to consider any backward compatibility that is not relevant to ERB or any template engines. So proposal (2) should be possible in ERB::Util.html_escape
unlike CGI.escapeHTML
.
Benchmark¶
Implementation: https://github.com/ruby/erb/pull/27
require 'benchmark/ips'
require 'erb'
class << ERB::Util
def html_escape_old(s)
CGI.escapeHTML(s.to_s)
end
end
Benchmark.ips do |x|
s = 'hello world'
x.report('before') { ERB::Util.html_escape_old(s) }
x.report('after') { ERB::Util.html_escape(s) }
x.compare!
end
ruby 3.1.2p20 (2022-04-12 revision 4491bb740a) [x86_64-linux]
Warming up --------------------------------------
before 1.066M i/100ms
after 1.879M i/100ms
Calculating -------------------------------------
before 10.615M (± 0.3%) i/s - 53.320M in 5.023083s
after 18.742M (± 0.4%) i/s - 93.929M in 5.011847s
Comparison:
after: 18741747.6 i/s
before: 10615137.1 i/s - 1.77x (± 0.00) slower