Thread: [PATCH] Tiny optmization or a bug?

[PATCH] Tiny optmization or a bug?

From
Ranier Vilela
Date:
Hi,
Hi,
Maybe this is a real bug.

The assignment has no effect, or forget dereferencing it?

Best regards.
Ranier Vilela

--- \dll\postgresql-12.0\a\backend\optimizer\plan\initsplan.c    Mon Sep 30 17:06:55 2019
+++ initsplan.c    Fri Nov 22 19:48:42 2019
@@ -1718,7 +1718,7 @@
                     relids =
                         get_relids_in_jointree((Node *) root->parse->jointree,
                                                false);
-                    qualscope = bms_copy(relids);
+                    bms_copy(relids);
                 }
             }
         }

Attachment

Re: [PATCH] Tiny optmization or a bug?

From
Tomas Vondra
Date:
On Fri, Nov 22, 2019 at 11:06:53PM +0000, Ranier Vilela wrote:
>Hi,
>Hi,
>Maybe this is a real bug.
>
>The assignment has no effect, or forget dereferencing it?
>
>Best regards.
>Ranier Vilela
>
>--- \dll\postgresql-12.0\a\backend\optimizer\plan\initsplan.c    Mon Sep 30 17:06:55 2019
>+++ initsplan.c    Fri Nov 22 19:48:42 2019
>@@ -1718,7 +1718,7 @@
>                     relids =
>                         get_relids_in_jointree((Node *) root->parse->jointree,
>                                                false);
>-                    qualscope = bms_copy(relids);
>+                    bms_copy(relids);
>                 }
>             }
>         }

Seriously, how are you searching for those "issues"?

1) We're using qualscope in an assert about 100 lines down, and as coded
we need a copy of relids because that may be mutated (and reallocated to 
a different pointer). So no, the assignment *has* effect.

2) bms_copy(relids) on it's own is nonsensical, because it allocates a
copy but just throws the pointer away (why making the copy at all).

Have you tried modifying this code and running the regression tests? If
not, try it.

   $ ./configure --enable-cassert
   $ make
   $ make check

Please, consider the suggestions from my previous response ...

regards

-- 
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services



RE: [PATCH] Tiny optmization or a bug?

From
Ranier Vilela
Date:
Hi,
I just wanted to help a little bit, sorry for the out balls.

Maybe, I got one or two right.

Anyway, thank you very much for your attention and patience.

best regards.
Ranier Vilela

________________________________________
De: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Enviado: sexta-feira, 22 de novembro de 2019 23:25
Para: Ranier Vilela
Cc: pgsql-hackers@postgresql.org
Assunto: Re: [PATCH] Tiny optmization or a bug?

On Fri, Nov 22, 2019 at 11:06:53PM +0000, Ranier Vilela wrote:
>Hi,
>Hi,
>Maybe this is a real bug.
>
>The assignment has no effect, or forget dereferencing it?
>
>Best regards.
>Ranier Vilela
>
>--- \dll\postgresql-12.0\a\backend\optimizer\plan\initsplan.c  Mon Sep 30 17:06:55 2019
>+++ initsplan.c        Fri Nov 22 19:48:42 2019
>@@ -1718,7 +1718,7 @@
>                                       relids =
>                                               get_relids_in_jointree((Node *) root->parse->jointree,
>                                                                                          false);
>-                                      qualscope = bms_copy(relids);
>+                                      bms_copy(relids);
>                               }
>                       }
>               }

Seriously, how are you searching for those "issues"?

1) We're using qualscope in an assert about 100 lines down, and as coded
we need a copy of relids because that may be mutated (and reallocated to
a different pointer). So no, the assignment *has* effect.

2) bms_copy(relids) on it's own is nonsensical, because it allocates a
copy but just throws the pointer away (why making the copy at all).

Have you tried modifying this code and running the regression tests? If
not, try it.

   $ ./configure --enable-cassert
   $ make
   $ make check

Please, consider the suggestions from my previous response ...

regards

--
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services