On Fri, 10 Feb 2012, Graycode wrote:

> I'm sending a patch to Philip instead of posting it here in order to
> enable him to review and adjust whatever he wants.

... which I will do in due course (but probably not for several weeks). 
The performance enhancement looks impressive.

> 1) I defined a new item in CONFIG.H to specify whether to retain the
> previous frame allocation behavior or to use this new method.  Perhaps
> Philip will choose not to have an option of selecting each method.

I agree with your approach for testing, but I don't think having an 
option is useful for a production version.

> 2) A way was need to communicate the frame basis between pcre_exec()
> and match() functions.  While a new parameter could have been used,
> I chose instead to add a (void *) pointer to the definition of the
> "match_data" structure declared in pcre_internal.h.

Good! Adding a new parameter to match() increases the amount of stack 
that is used when NO_RECURSE is *not* set (i.e. the case you are not 
interested in). It is something I really, really, really do not want 
ever to do. There are already some fudged cases using a similar 
technique to yours.

Philip

-- 
Philip Hazel

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

Reply via email to