On Mon, 29 Sep 2014, Nils Goroll wrote:
> https://www.varnish-cache.org/ uses pcre by compiling a domain language (VCL)
> into C and I wonder if we can offer varnish users a way of dynamically enabling
> JIT per RE.
>
> Obiously one could add some flag outside the RE, but I wonder if it would make
> sense to add an option setting (like "(*JIT)") the the RE itself, which would
> have the same effect as PCRE_STUDY_JIT_COMPILE when pcre_study() is called.
>
> There could be a PCRE_STUDY_JIT_OPTION flag to allow pcre_study to turn on JIT
> dynamically. Callers would have to oblige to calling pcre_free_study() when
> using PCRE_STUDY_JIT_OPTION.
This is something that could be thought about in the context of the new
API for PCRE that is being developed (known as PCRE2). Preliminary code
is already available for brave people who like testing things at
svn co svn://vcs.exim.org/pcre2/code/trunk pcre2
I am currently working on the documentation. Interpretive matching is
working, but JIT has yet to be implemented (though I have just
documented it :-).
In the new API, there is no longer an explicit "study" function - as it
turned out to be cheap, it now always done. For JIT there will be a new
function called pcre2_jit_compile(), to be called after pcre2_compile()
has been successful. Freeing of the JIT memory will happen automatically
when pcre2_code_free() is called.
It would be relatively simple to add support for (*JIT) to PCRE2. It is
unlikely that new developments such as this will now be added to PCRE1.
However, I am slightly wondering why you need to offer the option to
varnish users rather than just enabling JIT always, in order to get the
fastest performance. What have I missed?
Philip
--
Philip Hazel
