Re: [xml] libxml2 performance

["Peter Jacobi" <pj walter-graphtek com>]

Hi Bjorn, All,

Thank you for your detailed and informative profiling data.

For the second example, I'm unable to interpret the
line based data in some places. For example it attributes
2.85% to 'ctxt->token = 0' in a block only entered
when ctxt->token == '&' - but ctxt->token seems to
be unused, at least outside the DTD (and svg3 doesn't
have any '&').

Given that uncertainity, the data may imply, that dropping all
the ctxt->token checks may help.

Regarding the malloc cost, you would even come nearer
to the unfortunate 40% of my tests, if you use a version
of libxml-api.xml converted to data in elements 
(and converted to UTF8).

Regards,
Peter Jacobi