Fix oversight in handling of row-comparison index keys: if the row comparison
doesn't exactly match the index, we may have to change our initial positioning
strategy. For example, given an index on (f1,f2,f3) and a WHERE condition
"ROW(f1,f3) > ROW(2,3)", the code extracted the initial-positioning condition
"f1 > 2", which is wrong ... it has to be "f1 >= 2", else some rows matching
the WHERE condition may fail to be returned.
Applying patch to 8.2 only --- I'll fix it in HEAD later as part of the
planned index improvements (reverse-sort and NULLS FIRST/LAST work).
nbtsearch.c (r1.107 -> r126.96.36.199)
pgsql-committers by date
|Next:||From: Tom Lane||Date: 2007-01-07 03:00:12|
|Subject: Re: [COMMITTERS] pgsql: Check for ERANGE in exp() |
|Previous:||From: Peter Eisentraut||Date: 2007-01-07 00:13:55|
|Subject: pgsql: Allow XML fragment to contain a XML declaration.|