Hah that's a very good question @tevador

I haven't thought about that

How was it handled when RingCT was rolled out?

a seraphis enote can use ringct outputs as inputs, right? just as it can use seraphis enotes as inputs. for a time, could there be a blend of inputs allowed?

sorry, i think the proper phraseology would be "a seraphis tx can use ringct outputs as inputs"

Yes

There was no such issue with rct. A pre-rct output can be used as a rct output with a known mask.

And it looks like the same applies for seraphis, given hte few lines above ?

Can it also use pre-rct outputs the same way ?

moneromooo: that’s the plan

UkoeHB: CMIIW, but the Seraphis composition proof cannot be constructed for RingCT outputs due to a different key format, so you cannot have RingCT decoys when spending a Seraphis output

well, technically you can have RingCT decoys, but it's clear they are just decoys, so they are useless

tevador: I don't think we ever applied such an approach (e.g. the output pool was reset for RingCT)

And I think for Bulletproofs as well

Yes, a post-Seraphis TX input can be either CLSAG with all RingCT ring members or Seraphis with all Seraphis ring members. Outputs are always Seraphis outputs.

The point was that you cannot mix the decoys, so we need a new DSA for RingCT outputs that will be spent post-Seraphis.

Right

I *think* that is implicit for pre-rct. The pick is based on output index, so if there are very few new pre-rct outs, the picking automatically "skews" into the past.

I checked the code. We use a triangular distribution for pre-RCT outputs placed at the last existing output of the given value.

One option would be to use an approximate gamma tail distribution: johndcook.com/blog/2021/11/06/gamma-tail-estimate