>  | > | Changes you commited to the HEAD related to this PR seems good to me,
>  | > | everything works correctly and much faster than gawk (for huge
>  | > | regexps) that i used for years.
>  | > | 
>  | 
>  | unfortunately, there seem to be some bugs introduced by these commits. 
>  | I'm getting lots of junk pointer complaints from free() followed by a 
>  | segfault.  nawk prior to these changes don't have this problem.  The 
>  | patch Christos attached to PR/3406 makes the segfaults go away, but this 
>  | new nawk uses more than 10x the RAM on the same script with the same 
>  | input file than an older (prior to these changes) nawk.
>  | 
>  | I can send the awk script and input file if anyone cares to dig around.
>  
>  I fixed all of it...
>  christos

Are you sure about the following?

@@ -82,10 +82,10 @@
        void *p;
        int i, new_count;

-       if (state < fa->state_count)
+       if (++state < fa->state_count)
             return

I think ++ is unwanted

 new_count = state + 10;
 for (i = fa->state_count; i < new_count; ++i) {
                           ^^^^^
    ...
 }
 fa->state_count = new_count;

-- 
Best regards, Aleksey Cheusov.