Re: [PATCH] Resolve unknown-type literals in GRAPH_TABLE COLUMNS

On Mon, May 4, 2026 at 8:42 PM Peter Eisentraut <peter(at)eisentraut(dot)org> wrote:
>
> On 29.04.26 16:09, Ashutosh Bapat wrote:
> > On Mon, Apr 27, 2026 at 11:34 AM SATYANARAYANA NARLAPURAM
> > <satyanarlapuram(at)gmail(dot)com> wrote:
> >>
> >> Hi,
> >>
> >> On Sun, Apr 26, 2026 at 8:10 AM Junwang Zhao <zhjwpku(at)gmail(dot)com> wrote:
> >>>
> >>> Hi SATYANARAYANA,
> >>>
> >>> On Sun, Apr 26, 2026 at 3:53 AM SATYANARAYANA NARLAPURAM
> >>> <satyanarlapuram(at)gmail(dot)com> wrote:
> >>>>
> >>>> Hi hackers,
> >>>>
> >>>> transformRangeGraphTable() calls transformExpr() and
> >>>> assign_list_collations() for COLUMNS expressions but missed calling
> >>>> resolveTargetListUnknowns(). As a result, literals such as 'val1'
> >>>> in a COLUMNS clause retained type "unknown", causing failures with
> >>>> ORDER BY, UNION, and output conversions.
> >>>>
> >>>> Fix by calling resolveTargetListUnknowns() on the columns target
> >>>> list right after assign_list_collations(), similar to SELECT target lists in
> >>>> transformSelectStmt().
> >>>>
> >>>> Attached a patch to fix this, which also includes test cases to reproduce.
> >>>
> >>> I can reproduce this and the patch fixes it.
> >>>
> >>> One question: why is resolveTargetListUnknowns called after
> >>> assign_list_collations?
> >>>
> >>> I'm asking because in transformSelectStmt, resolveTargetListUnknowns
> >>> is invoked before assign_query_collations. It might not matter, but keeping
> >>> the order consistent would be good for readers.
> >>
> >>
> >> Updated the patch. It should be before.
> >
> > Do we really need a test for ORDER BY on a literal column? I replaced
> > all the test queries with a single one which covers all the scenarios
> > covered by those queries.
> >
> > The patch needed to consider pstate->p_resolve_unknowns. Unknown
> > literals are not resolved as text always. See the test query.
>
> I couldn't find a commit to apply this patch cleanly (the subject says
> patch 5/5, so maybe you had some unpublished local changes?).

I am maintaining all the SQL/PGQ fixes in the same branch and creating
patches from that branch. Hence 5/5. But still it shouldn't have
applied cleanly. Both git cherry-pick and git am <attached patch
file>, on a fresh branch succeeded. Please let me know if you still
face the issue.

> After
> applying the test case manually, it looks like the test output is
> already correct without the code change. So if this patch is still
> required, we need a better test case.
>

Yes, the test query doesn't reproduce the bug. Actually all but only
two queries from the Satya's patch show the bug. I have included a
test query which fails, with expected error message, without code
changes now. Also I don't think we need a separate section for this
test query. Included the query in one of the earliest sections in the
file.

--
Best Wishes,
Ashutosh Bapat