Re: Schema variables - new implementation for Postgres 15

On Sat, Aug 20, 2022 at 03:55:07PM +0200, Erik Rijkers wrote:
>
> Op 20-08-2022 om 15:41 schreef Pavel Stehule:
> > so 20. 8. 2022 v 15:36 odesílatel Erik Rijkers <er@xs4all.nl> napsal:
> >
> > > Op 20-08-2022 om 15:32 schreef Erik Rijkers:
> > > > Op 19-08-2022 om 17:29 schreef Pavel Stehule:
> > > >
> > > > make check  fails as a result of the errors in the attached
> > > > session_variables.out.
> > > >
> > >
> > >
> > > Sorry, that should have been this diffs file, of course (attached).
> > >
> >
> > It looks like some problem with not well initialized memory, but I have no
> > idea how it is possible. What are your configure options?
> >
>
> I compiled both assert-enable and 'normal', and I only just noticed that the
> assert-enable one did pass tests normally.
>
>
> Below is the config that produced the failing tests:
>
> ./configure
> --prefix=/home/aardvark/pg_stuff/pg_installations/pgsql.schema_variables
--bindir=/home/aardvark/pg_stuff/pg_installations/pgsql.schema_variables/bin.fast
--libdir=/home/aardvark/pg_stuff/pg_installations/pgsql.schema_variables/lib.fast
> --with-pgport=6986 --quiet --enable-depend --with-openssl --with-perl
> --with-libxml --with-libxslt --with-zlib  --enable-tap-tests
> --with-extra-version=_0820_schema_variables_1509    --with-lz4  --with-icu

I also tried locally (didn't look at the patch yet), with debug/assert enabled,
and had similar error:

diff -dU10 /Users/rjuju/git/postgresql/src/test/regress/expected/session_variables.out
/Users/rjuju/git/pg/pgmaster_debug/src/test/regress/results/session_variables.out
--- /Users/rjuju/git/postgresql/src/test/regress/expected/session_variables.out    2022-08-20 22:25:17.000000000 +0800
+++ /Users/rjuju/git/pg/pgmaster_debug/src/test/regress/results/session_variables.out    2022-08-20 22:30:50.000000000
+0800
@@ -983,23 +983,23 @@
 -- should to fail
 SELECT public.svar;
   svar
 ---------
  (10,20)
 (1 row)

 ALTER TYPE public.svar_test_type ADD ATTRIBUTE c int;
 -- should to fail too (different type, different generation number);
 SELECT public.svar;
-   svar
-----------
- (10,20,)
+        svar
+--------------------
+ (10,20,2139062142)
 (1 row)

 LET public.svar = ROW(10,20,30);
 -- should be ok again for new value
 SELECT public.svar;
     svar
 ------------
  (10,20,30)
 (1 row)

@@ -1104,31 +1104,31 @@
 (1 row)

 DROP VARIABLE public.svar;
 DROP TYPE public.svar_test_type;
 CREATE TYPE public.svar_test_type AS (a int, b int);
 CREATE VARIABLE public.svar AS public.svar_test_type;
 CREATE VARIABLE public.svar2 AS public.svar_test_type;
 LET public.svar = (10, 20);
 ALTER TYPE public.svar_test_type ADD ATTRIBUTE c int;
 SELECT public.svar;
-   svar
-----------
- (10,20,)
+    svar
+------------
+ (10,20,16)
 (1 row)

 LET public.svar2 = (10, 20, 30);
 ALTER TYPE public.svar_test_type DROP ATTRIBUTE b;
 SELECT public.svar;
- svar
--------
- (10,)
+  svar
+---------
+ (10,16)
 (1 row)

 SELECT public.svar2;
   svar2
 ---------
  (10,30)
 (1 row)