bug#17081: [PATCH] dfa: avoid undefined behavior

[Paul Eggert]

Paolo Bonzini wrote:
         \
> > +        size_t nbytes = mbrtowc (&_wc, lexptr, lexleft, &mbs); \
> > +        bool valid_char = 1 <= nbytes && nbytes < (size_t) -2; \
>
> I find these conditionals complicated to follow.

Yes, that identifier 'valid_char' was a confusing choice; as you noted, 
the character is valid even when nbytes is zero.

> I believe you should have simply
>     bool valid_char = nbytes < (size_t) -2;
>
> or better:
>
> > +        if (! valid_char)            \
>
>     if (nbytes >= (size_t) -2)

That wouldn't do, because when mbrtowc returns 0 the caller still needs 
to advance the pointer by 1 to get past the null byte, just as it needs 
to advance by 1 if mbrtowc returns (size_t) -2 or (size_t) -1.

> I see this patch has been committed already.  Can you please submit a followup?

There was a followup patch, in commit 2b9c57c, and the code's changed so 
that it no longer has a 'valid_char' local.  Perhaps it's clear enough now.