Re: [PATCH 3/4] dfa: refactor to prepare for upcoming optimizations

[Jim Meyering]

Paolo Bonzini wrote:
> * src/dfa.c (parse_bracket_exp): Move optimization of MBCSET from here...
> (addtok): ... to here.
> ---
>  src/dfa.c |   25 ++++++++++++++++---------
>  1 files changed, 16 insertions(+), 9 deletions(-)
>
> diff --git a/src/dfa.c b/src/dfa.c
> index 6602ae8..8fc6ed0 100644
> --- a/src/dfa.c
> +++ b/src/dfa.c
> @@ -1052,14 +1052,7 @@ parse_bracket_exp (void)
>      dfawarn (_("character class syntax is [[:space:]], not [:space:]"));
>
>  #if MBS_SUPPORT
> -  if (MB_CUR_MAX > 1
> -      && (!using_utf8()
> -          || invert
> -          || work_mbc->nchars != 0
> -          || work_mbc->nch_classes != 0
> -          || work_mbc->nranges != 0
> -          || work_mbc->nequivs != 0
> -          || work_mbc->ncoll_elems != 0))
> +  if (MB_CUR_MAX > 1)
>      {
>        static charclass zeroclass;
>        work_mbc->invert = invert;
> @@ -1463,7 +1456,21 @@ addtok (token t)
>  {
>  #if MBS_SUPPORT
>    if (MB_CUR_MAX > 1 && t == MBCSET)
> -    addtok_mb (MBCSET, ((dfa->nmbcsets - 1) << 2) + 3);
> +    {
> +      struct mb_char_classes *work_mbc = &dfa->mbcsets[dfa->nmbcsets - 1];

It's slightly tighter and more readable to make this temporary "const".
Only down-side is that inserting "const " pushes line-length to 80 ;-)

ACK either way.

> +      /* UTF-8 allows treating a simple, non-inverted MBCSET like a CSET.  */
> +      if (work_mbc->invert
> +          || (!using_utf8() && work_mbc->cset != -1)
> +          || work_mbc->nchars != 0
> +          || work_mbc->nch_classes != 0
> +          || work_mbc->nranges != 0
> +          || work_mbc->nequivs != 0
> +          || work_mbc->ncoll_elems != 0)
> +        addtok_mb (MBCSET, ((dfa->nmbcsets - 1) << 2) + 3);
> +      else
> +        addtok (CSET + work_mbc->cset);
> +    }
>    else
>  #endif
>      addtok_mb (t, 3);