Re: [pcre-dev] Posix API and JIT

On 11/16/2011 11:40 AM, Philip Hazel wrote:
> On Wed, 16 Nov 2011, Sheri wrote:
>
>> Here's a minimal request, please let me know if feasible.
>>
>> Suppose you added a regexec2 to the POSIX interface which passed study data
>> instead of NULL to pcre_exec. Couldn't the native pcre_study be used to create
>> study data based on the compiled pattern that was regcomp'ed? Mixed interface
>> for our older "global" regex'es?
> Performance would be poor because every time you called regexec2, it
> would have to do the study and the free ... and JIT study takes
> appreciative resources. So unless you only ever match a pattern exactly
> once, the performance gain might (would?) not be much, if anything. We
> already know that for relatively simple patterns matched once, JIT is
> not worth using. It comes into its own when you can do *one* JIT study,
> followed by *many* calls to pcre_exec.
>
> OK, now you are going to say why not do the study outside regexec, based
> on the POSIX data structure generated by regcomp (i.e. invent regstudy).
> If you do that, you have to pass it to regexec2 somehow ... this is
> changing your code quite a lot: adding a call to regstudy, modifying
> the regexec calls. I would argue that this is almost as much work as
> changing to the native API.
>
> I appreciate your situation, not a nice place to be in.
>
> Philip
>

Thanks, I'll take your word for it. I thought the structure for a
pattern compiled with regcomp was the same as one compiled with
pcre_compile and could be passed as-is to pcre_study. Oh well. There are
no external options used with the patterns if that makes any difference.

The change on our side as I had pictured it was to add (inside our
global matchG and replaceG functions) a call to pcre_study before
repeated calls to regexec2 (with arg specified for extra data from
pcre_study), eventually followed by existing regfree and added
pcre_study_free.

Regards,
Sheri