Hi,

On 2021-04-21 17:38:53 -0400, Tom Lane wrote:

 Andres Freund <andres@anarazel.de> writes:
 > I don't really see us getting rid of something like ctid as a generic
 > concept across AMs - there's just too many places that need a way to
 > reference a specific tuple. However, I think we ought to change how much
 > code outside of AMs know about what tids mean. And, although that's a
 > significant lift on its own, we ought to make at least the generic
 > representation variable width.

 It seems like it might not be that hard to convert ctid generically
 into a uint64, where heaps and heap-related indexes only use 6 bytes
 of it.

Yep.

 

 Variable-width I agree would be a very big complication added on top,
 and I'm not quite convinced that we need it.

I can see three (related) major cases where variable width tids would be
quite useful:
1) Creating an index-oriented-table AM would harder/more
   limited with just an 8 byte tid
2) Supporting "indirect" indexes (i.e. indexes pointing to a primary
   key, thereby being much cheaper to maintain when there are updates),
   would require the primary key to map to an 8 byte integer.
3) Global indexes would be a lot easier if we had variable width tids
   (but other ways of addressing the issue are possible).

Greetings,

Andres Freund