Re: tuple compare involving NULL

hi,

> This isn't really useful context, to me at least.

sorry to hear that.

> It tells me nothing of why you need an artificial PK

i don't understand what you mean with artificial. id _is_ the primary key.


 > or why you think a and b need to allow null.

they do. i don't control the database layout and i have to support
pagination involving NULLs in order by columns.


> The where clause is odd with its mix of q and t in the same row value

it is to allow ascending order on the first and descending order on the
second.


> and q.id is technically broken though I know this is just an
> example.

why is it? whenever (q.a, q.b) = (t.a, t.b) you need another (unique)
key to seek to the next row. maybe i am missing things.


> As muchas arbitrary default values suck they are at least better
> than null in this regard.  You can use coalesce for any data type.

is there a special value (of every type) that is minimal for ever
comparison (except with itself)? if so i could easily compare
  (COALESCE(t.a, MINIMAL VALUE), COALESCE(q.b, MINIMAL VALUE), t.id)

thank you for your time,
  tobias florek