Tgl wrote:
> [ shrug... ] Since userlocks are only advisory, a non-cooperating
> client can break anything in sight anyway. I don't find the above
> argument convincing. But in any case, you can use an OID or serial
> sequence identifier if you prefer that to CTID. They're just integers
> and it's really up to the user application to define the
interpretation
> of a userlock tag.
Right. My point is that use of CTIDs just too easy to screw up from the
user's perspective, because they change with updates. I was briefly
toying with a 'auto lock' mode which built the lock from ctid +
tableoid.
So, I'd suggest discouraging the use of ctid, just like the use of OIDs
is discouraged (in fact there is already a disclaimer about using ctid
as a logical identifier in the docs). Certainly we can't provide a
tighter integration between the user locks and the system columns via
extensions to the grammar, etc.
So it becomes a documentation issue.
Merlin
