Re: BUG #10189: Limit in 9.3.4 no longer works when ordering using a composite multi-type index

Hi,

I'm a coworker of Nick's and I was finally able to create a standalone test
case. I've attached a Java file Test.java
<http://postgresql.1045698.n5.nabble.com/file/n5802595/Test.java> . All you
have to do is change the first 3 lines of the main method to set your url,
username, and password, and make sure to include a JDBC jar. This
application will do the following

1. Create tables T1 and T2.
2. Begin populating rows into T1 and T2 and updating them in a "specific"
order
3. Every 1000 messages, a reindex will be attempted on T1's primary key.
This will fail eventually.

I've attached a screenshot of what you'll likely see as the console output.
<http://postgresql.1045698.n5.nabble.com/file/n5802595/error.png>

Here are some additional notes I have on the matter

1. I was able to reproduce the problem on both Windows and Linux with
Postgres 9.3.4.
2. I was able to reproduce the problem with the latest 9.3 JDBC jars
(although we are still using 9.1 for our software)
3. After applying the patch that you mentioned, I can no longer reproduce
the problem at all. It seems very likely that
http://git.postgresql.org/gitweb/?p=postgresql.git&a=commitdiff&h=c0bd128c8
<http://git.postgresql.org/gitweb/?p=postgresql.git&a=commitdiff&h=c0bd128c8>
is in fact the problem that we were seeing, but I'm attaching this test case
anyway just in case you can infer anything else from it.

How safe would it be if we took the 9.3.4 source code and just applied the
patch above. Like Nick said, we aren't aware of exactly what this patch does
and whether there are any commit dependencies it relies on.

Thanks for all your replies,
Wayne Huang

--
View this message in context: http://postgresql.1045698.n5.nabble.com/BUG-10189-Limit-in-9-3-4-no-longer-works-when-ordering-using-a-composite-multi-type-index-tp5802079p5802595.html
Sent from the PostgreSQL - bugs mailing list archive at Nabble.com.