On Mon, Nov 15, 2021 at 03:24:41PM -0800, Paul Eggert wrote:
> On 11/15/21 12:49, Carlo Marcelo Arenas Belón wrote:
> 
> > Apologies, I realize it is difficult to talk about code in abstract when
> > not inlined, but I think it will better addressed by "fixing" it as shown
> > in the attached patch.
> 
> That patch isn't right, because the relevant code inside libpcre2 uses
> size_t, not PCRE2_SIZE and that means our hacky workaround should use
> SIZE_MAX, not PCRE2_SIZE_MAX, so that it's consistent with libpcre2's
> internals.

You are correct, got confused by the changes going to PCRE2_SIZE -> idx, and
should had check, thanks for catching this.

Hopefully the next patch (which I was planning to do with some of the
cleanup you beat me to), will at least clear some possible future valgrind
complain.

> Of course this is all academic on all existing platforms, since IDX_MAX is
> way less than SIZE_MAX.

Agree, and frankly I still think it will be better if IDX_MAX would be used
unconditionally instead.

Carlo