On Sun, 14 Aug 2011, Zoltan Herczeg wrote:

> I am thinking how this feature affects the jit compiler, since the machine 
> code
> is part of the PCRE study data, but it cannot be reloaded (because pointers 
> are
> directly encoded into the instruction stream). Maybe the pointer should be
> cleared before the study data is saved.

I had realized that there was an issue here, and was going to email you 
about it. I guessed that the JIT data was not saveable, so the 
documentation will have to state that clearly. As it is up to the user 
to do the saving/restoring, there is nothing PCRE can do about clearing 
the pointer. I guess we say something like this: "If you are using the 
JIT optimization, note that the JIT study data cannot be saved and
restored. You will have to re-study the compiled pattern after reloading
if you want to recreate the JIT data." I am also planning on providing
an info call so that the user can tell whether JIT-study was successful.

In practice, I don't think many people actually use the save/restore 
facility.

I have had no feedback on the RC (except from you). I have other things 
to do for some of tomorrow, but as soon as I have time (maybe tomorrow 
afternoon, maybe Tuesday morning) I will make the proper 8.13 release 
and put it out. Then, as soon as you send me the corresponding JIT 
patches (or full code; I can do my own compares :-) I will start on the 
integration ... before people start reporting the next lot of bugs. :-(

Philip

-- 
Philip Hazel

-- 
## List details at https://lists.exim.org/mailman/listinfo/pcre-dev 

Reply via email to