Release 3.21.0: defer Bifrost-only commits to parity release
Reverts five Bifrost-only commits from the 3.21.0 mainnet cut so public mainnet matches the private 3.21.0 release, which does not carry them. They are deferred to a later Bifrost-parity release (branch itz/priv-3.21-bifrost-parity holds the forward-port). Each is a single revert commit referencing the original sha. S12 (restore terminal XMR errata completion) is NOT reverted — it ships in both public and private 3.21.0.
Reverted (reverse topo order)
| Original | Summary |
|---|---|
ce1bd504c6 |
fix(xmr): S07 - enforce deployment signer contracts |
76a39a4274 |
fix(xmr): S08 - harden scanner birthday and reorg recovery |
d4f057932b |
fix(observer): S13 - prevent XMR quarantine from stalling shared errata processing |
bfc12e4929 |
bifrost: anchor UTXO consolidation keysign on a shared THORChain block boundary |
3f963eec8b |
fix(xmr): S06c - reject signer HTTP redirects |
All five are Bifrost/observer/XMR-only (no consensus state-machine change). The reverts applied cleanly with no conflicts.
Consensus parity check
Verified that no public 3.21.0 consensus change is lost by this revert: every 3.21.0 consensus symbol is present in the private 3.21.0 release, and the only public commits absent from private are release-mechanics (version/upgrade handlers/openapi, for which the private cut carries its own equivalents) plus exactly these five deferred Bifrost commits.
Note for @zlyzol
The public helper vaultSignerCount (in x/thorchain/manager_slasher_current.go) is not present in the private tree. Private's needsNewVault does the same job via a different, more-advanced shape — it returns (needVault, durableKeepVault) and derives the signer-majority / stale-observation decision with committed-nonce + supermajority-observation logic rather than the single-bool helper. This revert does not touch that path (it only removes the five Bifrost commits above), but flagging it so you can eyeball public's vaultSignerCount vs private's needsNewVault for behavioral parity.