Re: [htmltmpl] Free Htmltemplate: Java implementation of HTML::Template

[<hn...@xe...>]

Sam Tregar <sa...@tr...> writes:

> On Wed, 18 Feb 2004, Hrvoje Niksic wrote:
>
>> Allow me to announce the availability of a Java implementation of
>> HTML::Template.  The home page is at:
>>
>>     http://htmltemplate.inet.hr/
>
> Wow, great work!  I love the website and your excellent
> documentation.

Thanks!

> I do, however, have a bone to pick.  You claim your system is much
> faster than HTML::Template and HTML::Template::JIT:
>
>    Free htmltemplate is significantly faster than HTML::Template in
>    rendering pre-compiled templates. Even with the JIT in place,
>    which according to Sam speeds up template rendering by a factor
>    of 4 to 8, htmltemplate might still be as fast or faster.
>
> Can you back this up with numbers?

Good point, I really should back up such a claim with numbers.  I
originally didn't go out of my way to make Htmltemplate very fast, I
simply coded it the "obvious" way.  But then I did some benchmarks
and, to my surprise, found it to be a good deal faster than I
expected.  The benchmark was admittedly *very* simple-minded: take a
template, compile it once, and run the template transformation over
and over to see how many transformations per second I can squeeze.
The result was astounding 3000 renderings per second, which was
totally beyond my expectations.  I first thought that I did something
wrong, that the code wasn't even running, but adding the output showed
that I was indeed producing the intended output 3000 times in a row in
about 1 second.

If my memory serves me, the same kind of benchmark showed
HTML::Template performing about 300 renderings per second.  The
reasoning that follows is summed up by the above paragraph: if JIT
speeds up HTML::Template 8 times, Htmltemplate might still be faster.
But I really haven't checked it.

I guess I should reword that paragraph to make it clear that it's
based on speculation and not on actual measurements of
HTML::Template::JIT's performance.

> Java has improved a great deal, granted, but I'd be pretty shocked
> if it could beat the highly-optimized C code generated by
> HTML::Template::JIT.

The code generated by HTML::Template::JIT still contains calls to
Perl's library and object system, doesn't it?  To me that's quite
different from typical hand-written C code, which I'd expect to always
beat Java by a wide margin.

How do you benchmark HTML::Template::JIT?  I can try to repeat the
same benchmark with Free Htmltemplate to see if I still get the
speedup I got when I last compared them.