Re: Questionable code in handling of wordend in the regexp engine in regex-emacs.c

[Alan Mackenzie <acm@muc.de>]

Hello, Stefan.

Sorry about the delay in replying.

On Sat, Feb 23, 2019 at 18:15:55 -0500, Stefan Monnier wrote:
> > Primarily, there is an

> >     UPDATE_SYNTAX_TABLE (charpos);

> > before determining the syntax of the previous character, which seems OK.
> > Later on, before determining the syntax of the next character, we have:

> >     UPDATE_SYNTAX_TABLE_FORWARD (charpos);

> > .  Between these two calls, charpos hasn't been changed.

> Good spotting.

Thanks!

> > Surely the argument to the second occurrence should be (charpos + 1)?

> I believe it's instead the other one that needs to use "charpos - 1"
> because the UPDATE_SYNTAX_TABLE is called just before reading the char
> *before* charpos (see patch below).

I don't think this is right.  offset is calculated from d, and then
decremented, before calculating charpos.

> > Also, probably less importantly, there is

> >     GET_CHAR_AFTER (c2, d, dummy);

> > , whereas at the same place in the handler for case symend: we have
> > instead

> >     c2 = RE_STRING_CHAR (d, target_multibyte);

> > .  Is the effect of these macros identical, or is one of them up to
> > date, and the other one really needs updating as well, for correct
> > functionality?

> According to my reading of the code, they're identical in multibyte
> buffers not in unibyte buffers where RE_STRING_CHAR just returns a value
> between 0 and 255 (i.e. ASCII or Latin-1 more or less), whereas
> GET_CHAR_AFTER will return either an ASCII char (0..127) or a raw-byte
> char (4194176..4194303).

OK.

> I think it's more correct to return a raw-byte char (4194176..4194303),
> so I'd tend to think that GET_CHAR_AFTER is the better choice, but
> please don't quote me on this.

I won't say a word!

> > I came across these whilst investigating bug #34525.  Making the
> > indicated changes to regex-emacs.c sadly doesn't help solve the symptoms
> > of that bug.  :-(

> Does the patch below help?

Unfortunately not, not for bug #34525.  I did try it out, though.  In the
mean time, I've advanced somewhat in the debugging.

>         Stefan


> diff --git a/src/regex-emacs.c b/src/regex-emacs.c
> index b667a43a37..72fb5ec561 100644
> --- a/src/regex-emacs.c
> +++ b/src/regex-emacs.c
> @@ -4813,7 +4813,7 @@ re_match_2_internal (struct re_pattern_buffer *bufp, re_char *string1,
>  	      int dummy;
 	      ptrdiff_t offset = PTR_TO_OFFSET (d) - 1;
>  	      ptrdiff_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
> -	      UPDATE_SYNTAX_TABLE (charpos);
> +	      UPDATE_SYNTAX_TABLE (charpos - 1);
>  	      GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
>  	      s1 = SYNTAX (c1);
 
-- 
Alan Mackenzie (Nuremberg, Germany).