2.9 Sprint Obsolete Markers Notes

Notes from the 2.9 development sprint pertaining to obsolete markers.

1. Cycles

o A''
|                 o A'
|        o  A     |
\        |        /
 \       |       /
  \      |      /
   \-----|-----/

A'' and A' both mark A obsolete, so you have a divergence. If Alice and Bob handle that case separately, Alice may pick A'' (by adding an obsolete marker for A'' -> A') and Bob may pick A' (by adding a marker for A' -> A''), thus introducing a direct cycle between A'' and A'.

Possible solutions for exchange, given we know the set of markers that should be on the destination:

sufficient = does the solution transmit everything MUST send - that is, the meta-history on both sides might mismatch, but the changeset history is correct on both sides

partial = can the solution transmit less than all the markers that have ever existed intelligently

box is "almost" sufficient for two reasons:

1. We can't push isolated kills

  o C ==x
  |
  o B
  |
  o A

• That is, if C is already on the server in draft phase, we can't push the kill marker for C with boxpush as it's currently defined. Proposal v1: push will inspect incoming draft-phase heads from the server. If any are dead locally, we push their kill marker to the server. We would exchange (pull or push):
• markers that apply to changesets being exchanged
• kill markers for ctx in common post-exchange Attempt two:

  We push [$NODE:↓:killed(heads(descendents($NODE)))]

• You can't handle a marker appearing after a push. That is

        α'       α
  o A'' == o A' == o A