Re: sync dfaexec from gawk: fixes real bugs in grep

[Paolo Bonzini]

> +  *end = eol;

As a follow-up I'd make this *end++ = eol, so that the code below either 
simplifies (fewer "+ 1") or becomes more idiomatic (e.g. comparing "< 
end" and ">= end" instead of <= and >).

> -      MALLOC(mblen_buf, unsigned char, end - (unsigned char const *)begin + 2);
> -      MALLOC(inputwcs, wchar_t, end - (unsigned char const *)begin + 2);
> +      MALLOC(mblen_buf, unsigned char, end - begin + 2);
> +      MALLOC(inputwcs, wchar_t, end - begin + 2);

Should this become + 3?

> +           if ((char *) p>  end)
> +             break;

So the sentinel doesn't work when doing MB matching?

> +      while ((t = trans[s]) != 0) { /* hand-optimized loop */

Would you change this comment to "hand-unrolled loop"?

> +      /* If the previous character was a newline, count it. */
> +      if (count&&  (char *) p<= end&&  p[-1] == eol)
> +       ++*count;

Probably present in gawk too but, why p <= end?  That would be always 
true due to the sentinel; wouldn't the right test be p > begin?

Or, what happens if you grep in a file that starts with '\n'?

> * src/search.c: Adjust to new dfaexec API.
> Now, dfaexec returns a pointer, not an integer,
> and the third parameter is END, not buffer size.

I suppose you'd squash 1/3 and 2/3 for bisectability?  Or could the 
search.c part be applied already to 1/3?

> +         *end = saved_end;                             \

Do we need to restore this or can we treat it as a scratch area?

Paolo