Horton

Thread

Commits

  1. Remove bogus Assert, add some regression test cases showing why.

  1. Failed Assertion about PolymorphicType

    Jaime Casanova <jaime.casanova@2ndquadrant.com> — 2020-04-04T21:03:52Z

    Hi,

Running sqlsmith on master i got an assertion failure on parse_coerce.c:2049

This is a minimal query to reproduce in an empty database, i also
attached the stack trace

 """
select
  pg_catalog.array_in(
    cast(pg_catalog.regoperatorout(
      cast(cast(null as regoperator) as regoperator)) as cstring),
    cast((select pronamespace from pg_catalog.pg_proc limit 1 offset 1)
       as oid),
    cast(subq_1.pid as int4)) as c0
from pg_catalog.pg_stat_progress_analyze as subq_1
 """

-- 
Jaime Casanova                      www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

  2. Re: Failed Assertion about PolymorphicType

    Tom Lane <tgl@sss.pgh.pa.us> — 2020-04-04T21:21:07Z

    Jaime Casanova <jaime.casanova@2ndquadrant.com> writes:
> Running sqlsmith on master i got an assertion failure on parse_coerce.c:2049

Hmph, or more simply:

regression=# select array_in('{1,2,3}',23,-1);
server closed the connection unexpectedly

which is a case that worked before.  The core of the problem is
that array_in() violates the assumption that a polymorphic result
requires a polymorphic argument:

regression=# \df array_in
                            List of functions
   Schema   |   Name   | Result data type |  Argument data types  | Type 
------------+----------+------------------+-----------------------+------
 pg_catalog | array_in | anyarray         | cstring, oid, integer | func
(1 row)

I see that enforce_generic_type_consistency did not use to assert
that it'd resolved every polymorphic rettype.  So I think we should just
remove that assertion (and fix the incorrect comment that led to
adding it).

			regards, tom lane