Large objects fixed

Well, I've found the problem that was breaking Large Objects and, although
I have a fix, I still believe the cause is still there.

It wasn't a protocol problem after all, but a memory leak that was causing
the backend to throw a Segmentation Violation when an argument was being
free'd.

For example:

Running src/test/example/testlo2, it calls lo_import on a file, then
lo_export to export the newly formed large object to another file.

Anyhow, every time, on the last call to lo_write (when importing the
file), the backend seemed to just die. The only difference between this
call and the previous calls, is that the amount of data to write is
smaller. Even changing the block size didn't change this fact.

Anyhow, I'm now trying to break it before posting the patch, but both
libpq & JDBC are running flawlessly.

Hopefully, the patch will be up later this afternoon.

--
Peter T Mount  petermount@earthling.net or pmount@maidast.demon.co.uk
Main Homepage: http://www.demon.co.uk/finder
Work Homepage: http://www.maidstone.gov.uk Work EMail: peter@maidstone.gov.uk