I’m sure you’re wondering what’s the reason for Hpricot updates, in
the face of heated competition from the Nokogiri and LibXML
libraries. Remember that Hpricot has no dependencies and is smaller
than either of those libs. Hpricot uses its own Ragel-based
parser, so you have the freedom to hack the parser itself, the code
is dwarven by comparison.
Best of all, Hpricot has run on JRuby in the past. And I am in the
process of merging some IronRuby code[1] and porting 0.7 to
JRuby. This means your code will run on a variety of Ruby platforms
without alteration. That alone makes it worthwhile, wouldn’t you
agree?
Clearly, the benchmarks you see on Ruby Inside are skewed to favor
Nokogiri. They parse XML through Hpricot without using Hpricot.XML(),
which is not only wrong, but puts XML through needless HTML cleanup
operations. I am sure that Hpricot 0.7 still fares slower on large
documents. However, for instance, try testing a large amount of
small documents (a much more common scenario) with this latest
version.
You have to question a benchmark that is entirely based on two XML
documents. What about HTML fix ups? What about various platforms
and CPUs? Why not treat Hpricot fairly and use it properly in the
benchmarks? It reeks of something.
I’m sure you’re wondering what’s the reason for Hpricot updates, in
the face of heated competition from the Nokogiri and LibXML
libraries. Remember that Hpricot has no dependencies and is smaller
than either of those libs. Hpricot uses its own Ragel-based
parser, so you have the freedom to hack the parser itself, the code
is dwarven by comparison.
Also, isn’t Hpricot more accepting of skanky HTML? m.
You have to question a benchmark that is entirely based on two XML
documents. What about HTML fix ups? What about various platforms
and CPUs? Why not treat Hpricot fairly and use it properly in the
benchmarks? It reeks of something.
Here’s what I use N for:
//form[
./descendant::fieldset[
./descendant::legend and
./descendant::li[
./descendant::label and
./descendant::input ]
]
]
I generate that from some N::HTML::Builder code, form{ fieldset { etc }
}, which
turns into a DOM containing etc . The
goal is
an assertion like this:
assert_xhtml do
h2 'Sales'
select! :size => SaleController::LIST_SIZE do
option names[1]
option names[0]
end
end
The point is to match an example HTML to a target HTML. I first tried it
by
walking that object model myself, recursing thru all DOM children to
find the
ones that match. However, as the recursion got more complex, I was
“adding
epicycles” to the code.
I backed off and rewrote, by first converting all the example HTML into
one
jiy-normous XPath, shown above. I have to do it like this because the
example
HTML could contain anything, and I need the query to run fast and
absolutely
stable. My assert_xhtml should not fail if the target code has the
correct HTML
subset - or vice versa. I can’t do that anywhere except LibXML, and I
need to
keep that easy to install.
And, in the grand scheme of things, I don’t think you have room to
complain
about your libraries’ adoption rates!
I’m sure you’re wondering what’s the reason for Hpricot updates, in
the face of heated competition from the Nokogiri and LibXML
libraries. Remember that Hpricot has no dependencies and is smaller
than either of those libs. Hpricot uses its own Ragel-based
parser, so you have the freedom to hack the parser itself, the code
is dwarven by comparison.
Also, isn’t Hpricot more accepting of skanky HTML?
no. I’ve had a bug open for years on hpricot because it couldn’t deal
with the relatively simple forms on the trackers on rubyforge.org.
nokogiri dealt with it perfectly and since mechanized migrated from
hpricot to nokogiri I’ve had fewer issues overall.
I should reemphasize… YEARS. Even the bug tracker has since
disappeared. This is where nokogiri really shines IMBO(*).
*) in my biased opinion. I work/hang out with aaron patterson on a
regular basis. That said, he fixes bugs I (and others–I watch) report
in a timely basis.
You have to question a benchmark that is entirely based on two XML
documents. What about HTML fix ups? What about various platforms
and CPUs? Why not treat Hpricot fairly and use it properly in the
benchmarks? It reeks of something.
You do have to question it (as you should question all benchmarks,
really)… But that question should come in the form of a bug report,
or a patch. To do otherwise… reeks of something.
Best of all, Hpricot has run on JRuby in the past. And I am in the
process of merging some IronRuby code[1] and porting 0.7 to
It seems like my port of Hpricot to IronRuby did not go unnoticed
despite having kept quiet about it so far
By the way, porting 0.7 to IronRuby is on my radar: I am just not sure
about how long this will take (I am pretty busy as of lately) but
staying up to date with the current latest version of Hpricot is
indeed something I want to achieve.
On Wed, Mar 18, 2009 at 03:08:39AM +0900, _why wrote:
than either of those libs. Hpricot uses its own Ragel-based
Nokogiri. They parse XML through Hpricot without using Hpricot.XML(),
which is not only wrong, but puts XML through needless HTML cleanup
operations. I am sure that Hpricot 0.7 still fares slower on large
documents. However, for instance, try testing a large amount of
small documents (a much more common scenario) with this latest
version.
Thank you for pointing out my mistakes. The repository[1] is public in
order to keep myself honest. Patches are welcome.
You have to question a benchmark that is entirely based on two XML
documents. What about HTML fix ups? What about various platforms
and CPUs? Why not treat Hpricot fairly and use it properly in the
benchmarks? It reeks of something.
HTML fix ups will be tested as well. So will CSS searches, XPath
searches, memory usage, and many other things. As I said[2], these
benchmarks
are not complete. If you’re worried about being treated fairly, fork my
repository and write tests.
You have to question a benchmark that is entirely based on two XML
documents. What about HTML fix ups? What about various platforms
and CPUs? Why not treat Hpricot fairly and use it properly in the
benchmarks? It reeks of something.
Don’t be an ass. Code (and benchmark results) speak much louder than
snark. Aaron has put the current benchmarks up on GitHub[1], and I’m
sure he’ll welcome any patches, additions, or corrections.
On Wed, Mar 18, 2009 at 06:56:19AM +0900, Aaron P. wrote:
HTML fix ups will be tested as well. So will CSS searches, XPath
searches, memory usage, and many other things. As I said[2], these benchmarks
are not complete. If you’re worried about being treated fairly, fork my
repository and write tests.
No no, don’t be silly, I’d much rather complain and be a sore
loser. I insist.
Look, I think I’d just rather see the benchmarks kept up by a
third party who has nothing to gain and can show a more nuanced
view of the scene. I really wish I could drop Hpricot (as
RubyfulSoup did,) but I think it has its strengths.
Let me ask you this. You’re neck and neck with libxml-ruby. The
bulk of your time is spent in the exact same HTML parser as
libxml-ruby. Why the hyperfocus on benchmarks and declaring
yourselves winners? You’re never going to be too far off from
their speed. So, I mean, it strikes me as adversarial and needless,
if your library quality and bug fixing are of the sort that Ryan
David has just touted.
On Wed, Mar 18, 2009 at 08:26:38AM +0900, _why wrote:
third party who has nothing to gain and can show a more nuanced
view of the scene. I really wish I could drop Hpricot (as
RubyfulSoup did,) but I think it has its strengths.
I agree, but who will write them? So far, we only have either poorly
written ones, or speculation. Neither are good. I figured if I wrote
these, put it on github, I could get other people to do the work.
Let me ask you this. You’re neck and neck with libxml-ruby. The
bulk of your time is spent in the exact same HTML parser as
libxml-ruby. Why the hyperfocus on benchmarks and declaring
yourselves winners? You’re never going to be too far off from
their speed. So, I mean, it strikes me as adversarial and needless,
if your library quality and bug fixing are of the sort that Ryan
David has just touted.
I’m not sure that 10-20% difference is neck and neck. I actually found
this result to be a surprise. I thought nokogiri would be slower. In
fact, I am sure I will find parts that are slower. Once I do, I know
where I can improve.
I don’t understand why you would think this is adversarial. As I said,
these benchmarks are not finished. I am merely trying to collect data,
and I want no emotions involved. I made my tests public so that
hopefully I can remain unbiased. If it seems unfair or incorrect, tell
me so. It won’t hurt my feelings. My goal is to learn, and to write
better software.
I don’t mind benchmarks as much as the constant cat-and-mouse game we
have to play. Ultimately most of the microbenchmarks published are
meaningless, but we have to spend a lot of time flexing that muscle to
remain a contender. It’s tiring
powerbook-g4-15-de-villa:/opt/local/bin villa$ sudo gem install hpricot
–source http://code.whytheluckystiff.net
Building native extensions. This could take a while…
Successfully installed hpricot-0.7
1 gem installed
Installing ri documentation for hpricot-0.7…
Installing RDoc documentation for hpricot-0.7…
powerbook-g4-15-de-villa:/opt/local/bin villa$ irb
irb(main):001:0> require ‘rubygems’
=> true
irb(main):002:0> require ‘hpricot’
LoadError: Failed to load
/usr/local/lib/ruby/gems/1.8/gems/hpricot-0.7/lib/hpricot_scan.bundle
from
/usr/local/lib/ruby/gems/1.8/gems/hpricot-0.7/lib/hpricot_scan.bundle
from
/usr/local/lib/ruby/site_ruby/1.8/rubygems/custom_require.rb:27:in require' from /usr/local/lib/ruby/gems/1.8/gems/hpricot-0.7/lib/hpricot.rb:20 from /usr/local/lib/ruby/site_ruby/1.8/rubygems/custom_require.rb:32:in gem_original_require’
from
/usr/local/lib/ruby/site_ruby/1.8/rubygems/custom_require.rb:32:in
`require’
from (irb):2
On Tue, Mar 17, 2009 at 10:21 PM, Charles Oliver N. [email protected] wrote:
You have to question a benchmark that is entirely based on two XML
documents. What about HTML fix ups? What about various platforms
and CPUs? Why not treat Hpricot fairly and use it properly in the
benchmarks? It reeks of something.
Welcome to my personal hell.
Ironic that Peter just posted a positive note about new JRuby benchmarks
and since mechanized migrated from hpricot
to nokogiri I’ve had fewer issues overall.
I have had issues after mechanize migrated to nokogiri. In fact I am
using the older mechanize without the dependency on nokogiri, until I am
able to install nokogiri without a problem.
(PS: For the record, I never use rubygems and never will for various
reason, most importantly because I do not need and do not want automatic
dependency handling without me controlling it, so a part of this issue
is surely my own doing. But fact remains that the older mechanize at the
moment works like a charm for me, whereas the newer mechanize does not
work because I can not install nokogiri easily. Just for the record, the
error with nokogiri “rake” is:
"3) Failure:
test_exslt(TestXsltTransforms) [./test/test_xslt_transforms.rb:76]:
<“2009-03-20”> expected to be =~
</\d{4}-\d\d-\d\d[-|+]\d\d:\d\d/>.
Since Mechanize can use either Nokogiri or Hpricot as a backend, it
seems like a good idea if neither were an actual dependency.
Actually, IMO they should both be alternative dependencies. Which, of
course, RubyGems doesn’t support. But since Marc doesn’t use RubyGems,
it should work fine.
