This is a specific use case, I have a big table without a pk. Updates with ctid are blazing fast even without an index. I dont need it.
The argument behind this is that users excpect this functionality, its not just me. Search stackoverflow. They end up using various suboptimal solutions as I described earlier. This is a very very simple functionality so please consider it. Im also writing an opensource lib that would make use of this. My users will be thankfull to you.
On Sat, Apr 17, 2021 at 12:58 PM Vladimír Houba ml. <v.houba@gmail.com> wrote:
I use ctid as a row identifier within a transaction in a Java application.
This doesn't present a very compelling argument since an actual user declared primary key is what is expected to be used as a row identifier. And as those are typically bigint if you follow this norm you get exactly what you say you need.
