Re: Making aggregate deserialization (and WAL receive) functions slightly faster

David Rowley <dgrowleyml(at)gmail(dot)com> writes:
> I spent more time on this and did end up with 2 new init functions as
> you mentioned. One for strictly read-only (initReadOnlyStringInfo),
> which cannot be appended to, and as you mentioned, another
> (initStringInfoFromString) which can accept a palloc'd buffer which
> becomes managed by the stringinfo code. I know these names aren't
> exactly as you mentioned. I'm open to adjusting still.

This v3 looks pretty decent, although I noted one significant error
and a few minor issues:

* in initStringInfoFromString, str->maxlen must be set to len+1 not len

* comment in exec_bind_message doesn't look like pgindent will like it

* same in record_recv, plus it has a misspelling "Initalize"

* in stringinfo.c, inclusion of pg_bitutils.h seems no longer needed

I guess the next question is whether we want to stop here or
try to relax the requirement about NUL-termination. I'd be inclined
to call that a separate issue deserving a separate commit, so maybe
we should go ahead and commit this much anyway.

regards, tom lane