The reasoning for it is documented with itself. The plan is to use it within
our header for committing to the DAG (allowing one header per epoch, yet
logarithmic proofs for any header within the epoch), the transactions
commitment (allowing logarithmic proofs of a transaction within a block,
without padding), and the events commitment (allowing logarithmic proofs of
unique events within a block, despite events not having a unique ID inherent).
This also defines transaction hashes and performs the necessary modifications
for transactions to be unique.
I believe this was originally here as we needed to return a reference, not an
owned instance, so this caching enabled returning a reference? Regardless, it
isn't valuable now.
This is read from the BABE pre-digest when converting from a SubstrateHeader.
This causes the genesis block to have time 0 and all blocks produced with BABE
to have a time of the slot time. While the slot time is in 6-second intervals
(due to our target block time), defining in milliseconds preserves the ABI for
long-term goals (sub-second blocks).
Usage of the slot time deduplicates this field with BABE, and leaves the only
possible manipulation to propose during a slot or to not propose during a slot.
The actual reason this was implemented this way is because the Header trait is
overly restrictive and doesn't allow definition with new fields. Even if we
wanted to express the timestamp within the SubstrateHeader, we can't without
replacing Header::new and making a variety of changes to the polkadot-sdk
accordingly. Those aren't worth it at this moment compared to the solution
implemented.
Consolidates all primitives into a single crate. We didn't benefit from its
fragmentation. I'm hesitant to say the new internal-organization is better (it
may be just as clunky), but it's at least in a single crate (not spread out
over micro-crates).
The ABI is the most distinct. We now entirely own it. Block header hashes don't
directly commit to any BABE data (avoiding potentially ~4 KB headers upon
session changes), and are hashed as borsh (a more widely used codec than
SCALE). There are still Substrate variants, using SCALE and with the BABE data,
but they're prunable from a protocol design perspective.
Defines a transaction as a Vec of Calls, allowing atomic operations.
