Questions about PITR

I have a couple of question about PITR. I believe some of the things that 
PITR must do may be applicable to materialized views and some other 
projects I've been thinking about.

I guess I should ask how PITR works. I imagine it does something like the 
following. Correct me if I am wrong. First, record every change to the 
database somewhere. When you go to restore a database, first load the last 
backup, and then reapply all the changes recorded since the last backup.

It's the step of recording the changes that I am interested in. I don't know 
the method of recording -- how to capture the changes, which changes to 
capture, and how to record them.

I'm writing some stuff for deferred updates materialized views. What I do is 
record any change to a particular table. Then when the materialized view is 
updated, I pick through the changes and discover which rows need to be 
updated in the MV.

I think it would be easier if I used the same code that PITR uses to record 
the changes. How probable is that? Also, the same code that records changes 
for PITR may be useful for replication, change logs on particular records, 
or any number of things.

-- 
Jonathan Gardner
jgardner@jonathangardner.net