Hi,
On 2019-05-21 19:38:57 +0200, David Fetter wrote:
> On Tue, May 21, 2019 at 12:32:21PM -0400, Robert Haas wrote:
> > Defaulting BUFFERS to ON is probably a reasonable change, thuogh.
>
> Would this be worth back-patching? I ask because adding it will cause
> fairly large (if mechanical) churn in the regression tests.
This is obviously a no. But I don't even know what large mechanical
churn you're talking about? There's not that many files with EXPLAIN
(ANALYZE) in the tests - we didn't have any until recently, when we
added SUMMARY OFF, to turn off non-deterministic details (f9b1a0dd4).
$ grep -irl 'summary off' src/test/regress/{sql,input}
src/test/regress/sql/select.sql
src/test/regress/sql/partition_prune.sql
src/test/regress/sql/tidscan.sql
src/test/regress/sql/subselect.sql
src/test/regress/sql/select_parallel.sql
adding a bunch of BUFFERS OFF to those wouldn't be particularly
painful. And if we decided it somehow were painful, we could infer it
from COSTS or such.
Greetings,
Andres Freund