- Title: fix-prior-fees
- Authors: John Tromp
- Start date: January 21, 2021
- RFC PR: mimblewimble/grin-rfcs#0021
- Tracking issue: mimblewimble/grin#3635
Carry the restriction of fees, to 40 bits since HF4, back to all history.
Do not change headerversion despite the hard-forking nature.
This makes the FeeFields methods fee() and fee_shift() height independent,
as well as several other functions which end up calling them.
This results in nontrivial code and consensus model simplification,
with no downside, which is always a good thing.
The largest fee to appear on mainnet prior to Hark Fork 4 is 2.404 Grin. This means that extending the current 40-bit fee restriction to the pre-HF4 history preserves validity. While this is technically a hard-forking change, we could treat it as an optional update. An attacker could in theory split the network by creating an ultra-deep reorg that includes a pre-HF4 fee exceeding 40 bits. However, the split damage would utterly pale in comparison to having a many months deep reorg in the first place. So in practice this change does not justify a header version bump and mandatory upgrade.
When computing the fee paid by a transaction, take all the fee fields in transaction kernels, and mask out all but the least significant 40 bits. This masking used to be conditional on the header version of a argument height being at least 5 [1].
None!
There is no compelling reason for leaving a height dependency in the fee method.
How many months should we wait after HF4 before rolling out this code simplification? I propose that 3 months is more than enough, making this RFC acceptable from mid April 2021 onwards.
[1] transaction.rs