> This entirely misses the network cost. Yes, sure, we can send
> "diffs", but if you send enough diffs eventually you send a lot of data.

The whole point of that section of the email was to consider the
network cost. There are many cases for which transmitting a
supplementary 1-in-1-out transaction (i.e. a sponsorship txn) is going
to be more efficient from a bandwidth standpoint than rebroadcasting a
potentially large txn during RBF.

> > In an ideal design, special structural foresight would not be
> > needed in order for a txn's feerate to be improved after broadcast.
> >
> > Anchor outputs specified solely for CPFP, which amount to many
> > bytes of wasted chainspace, are a hack. > It's probably
> > uncontroversial at this
>
> This has nothing to do with fee bumping, though, this is only solved
> with covenants or something in that direction, not different relay
> policy.

My post isn't only about relay policy; it's that txn
sponsors allows for fee-bumping in cases where RBF isn't possible and
CPFP would be wasteful, e.g. for a tree of precomputed vault

> How does this not also fail your above criteria of not wasting block
> space?

In certain cases (e.g. vault structures), using sponsorship txns to
bump fees as-needed is more blockspace-efficient than including
mostly-unused CPFP "anchor" outputs that pay to fee-management wallets.
I'm betting there are other similar cases where CPFP anchors are
included but not necessarily used, and amount to wasted blockspace.

> Further, this doesn't solve pinning attacks at all. In lightning we
> want to be able to *replace* something in the mempool (or see it
> confirm soon, but that assumes we know exactly what transaction is in
> "the" mempool). Just being able to sponsor something doesn't help if
> you don't know what that thing is.

When would you be trying to bump the fee on a transaction without
knowing what it is? Seeing a specific transaction "stuck" in the
mempool seems to be a prerequisite to bumping fees. I'm not sure what
you're getting at here.