Re: pgsql: Add more SQL/JSON constructor functions

On Tue, Jul 2, 2024 at 3:19 PM jian he <jian(dot)universality(at)gmail(dot)com> wrote:
> On Mon, Jul 1, 2024 at 6:45 PM Amit Langote <amitlangote09(at)gmail(dot)com> wrote:
> >
> > On Sun, Jun 30, 2024 at 3:56 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > > Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> writes:
> > > >> + /*
> > > >> + * For domains, consider the base type's typmod to decide whether to setup
> > > >> + * an implicit or explicit cast.
> > > >> + */
> > > >> + if (get_typtype(returning->typid) == TYPTYPE_DOMAIN)
> > > >> + (void) getBaseTypeAndTypmod(returning->typid, &baseTypmod);
> > >
> > > > TBH I'm not super clear on why we decide on explicit or implicit cast
> > > > based on presence of a typmod. Why isn't it better to always use an
> > > > implicit one?
> > >
> > > Hmm ... there are a bunch of existing places that seem to have similar
> > > logic, but they are all in new-ish SQL/JSON functionality, and I would
> > > not be surprised if they are all wrong. parse_coerce.c is quite
> > > opinionated about what a domain's typtypmod means (see comments in
> > > coerce_type() for instance); see also the logic in coerce_to_domain:
> > >
> > > * If the domain applies a typmod to its base type, build the appropriate
> > > * coercion step. Mark it implicit for display purposes, because we don't
> > > * want it shown separately by ruleutils.c; but the isExplicit flag passed
> > > * to the conversion function depends on the manner in which the domain
> > > * coercion is invoked, so that the semantics of implicit and explicit
> > > * coercion differ. (Is that really the behavior we want?)
> > >
> > > I don't think that this SQL/JSON behavior quite matches that.
> >
> > The reason I decided to go for the implicit cast only when there is a
> > typmod is that the behavior with COERCION_EXPLICIT is only problematic
> > when there's a typmod because of this code in
> > build_coercion_expression:
> >
> > if (nargs == 3)
> > {
> > /* Pass it a boolean isExplicit parameter, too */
> > cons = makeConst(BOOLOID,
> > -1,
> > InvalidOid,
> > sizeof(bool),
> > BoolGetDatum(ccontext == COERCION_EXPLICIT),
> > false,
> > true);
> >
> > args = lappend(args, cons);
> > }
> >
> > Yeah, we could have fixed that by always using COERCION_IMPLICIT for
> > SQL/JSON but, as Jian said, we don't have a bunch of casts that these
> > SQL/JSON functions need, which is why I guess we ended up with
> > COERCION_EXPLICIT here in the first place.
> >
> > One option I hadn't tried was using COERCION_ASSIGNMENT instead, which
> > seems to give coerceJsonFuncExpr() the casts it needs with the
> > behavior it wants, so how about applying the attached?
>
> you patched works.
> i think it's because of you mentioned build_coercion_expression ` if
> (nargs == 3)` related code
> and
>
> find_coercion_pathway:
> if (result == COERCION_PATH_NONE)
> {
> if (ccontext >= COERCION_ASSIGNMENT &&
> TypeCategory(targetTypeId) == TYPCATEGORY_STRING)
> result = COERCION_PATH_COERCEVIAIO;
> else if (ccontext >= COERCION_EXPLICIT &&
> TypeCategory(sourceTypeId) == TYPCATEGORY_STRING)
> result = COERCION_PATH_COERCEVIAIO;
> }
>
> functions: JSON_OBJECT,JSON_ARRAY, JSON_ARRAYAGG,JSON_OBJECTAGG,
> JSON_SERIALIZE
> the returning type can only be string type or json. json type already
> being handled in other code.
> so the targetTypeId category will be only TYPCATEGORY_STRING.

Yes, thanks for confirming that.

I checked other sites that use COERCION_ASSIGNMENT and I don't see a
reason why it can't be used in this context.

I'll push the patch tomorrow unless there are objections.

--
Thanks, Amit Langote