BUG #4491: regression in gist indexes

The following bug has been logged online:

Bug reference:      4491
Logged by:          Jeff Frost
Email address:      jeff@frostconsultingllc.com
PostgreSQL version: 8.3.4
Operating system:   Fedora 9/Gentoo/Mac OS X
Description:        regression in gist indexes
Details:

It seems that 8.3.4 has a regression in the updating of gist indexes.  After
updating an indexed column in a row with a postgis gist index, the row is no
longer found in an index scan.  I have verified that this works on 8.2.7,
8.3.0, 8.3.1 and 8.3.3 but not 8.3.4.
Verified this is a problem on Fedora 9/Gentoo/Mac OS X.

You can see the test case here:

https://gist.github.com/d6c9b183196717d73b6a

Interestingly, the first index scan after the update does find the row, but
subsequent scans do not.

Here you can see it working on 8.3.0:

https://gist.github.com/d0dbbf29606822b8ceb9

You can grab the test table data dump here:

http://www.frostconsultingllc.com/coordinate_test.dmp

Please contact me if there's anything else you need to reproduce the bug.
Note that if you're using the coordinate_test data, you'll have to set
enable_seqscan = 0 to force an index scan as the table only contains 100
rows.