"Merlin Moncure" <mmoncure@gmail.com> writes:
> On 5/19/06, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> This only affects the 7.4 and 8.0 branches, because earlier and later
>> versions of Postgres don't use this technique for detecting duplicates.
>> But it's surprising we didn't find it before.
> hm. about a year ago I reported a case where the database allowed
> multiple records with the same p-key which were causing problems with
> dump/reload from 8.0->8.1. It was pretty rare, but it looked like
> under certain circumstances unique constraint was not getting applied.
> I was unable to reproduce it, though.
Yeah, I remember.
> Is it possible that this bug was the cause of that particular problem?
No, this is unrelated. It only occurs in a query that's fetching rows
using OR'd indexscans, eg
SELECT ... WHERE indexedcol = 42 OR indexedcol = 47 OR ...;
(you can spell it "indexedcol IN (42,47,...)" with same results) and
the problem is basically incorrect detection of fetching the same row
more than once, ie, a bug in the code that's in charge of not returning
rows multiple times if query is like
SELECT ... WHERE indexedcol = 42 OR indexedcol = 42 OR ...;
This is nowhere near the unique-constraint enforcement mechanism.
regards, tom lane