This crate contains the data structures and core algorithms that comprise Mel's core state transition function. Any piece of software needing to validate Mel transactions or answer questions like "what happens to the Mel state if transactions A, B, and C happen" can use this minimal-dependency crate.
The most important type in the crate is State, and the closely associated type SealedState. The yellow paper talks about them further, but in short:
Staterepresents an mutable Mel world-state and it's not directly serializable. It includes all the information needed to validate new transactions and blocks, such as a SMT of all outstanding coins, Melmint parameters, etc. It has methods takingTransactions etc that advance the state, as well as others to produce serializable blocks, headers, etc.SealedStaterepresents a sealed state. This roughly corresponds to the notion of "the blockchain state at a given height". Blocks represent transitions from oneSealedStateto another.
This crate is the most consensus-critical part of Mel, and essentially defines the entire on-chain logic of the Mel blockchain.
Versions incompatible with the latest Mel state are thus all yanked.
To illustrate STF's usage, let's look at melnode as an example.
When a node is run for the first time, it will instantiate storage to keep track of the State. It does this by reading it from persisted history data, or create a new genesis state.
During it's lifetime, the node will need to update it's current knowledge of the State by doing things like handling incoming apply_tx RPCs, or syncing up its blocks (apply_block) with other nodes.
For example, an incoming apply_tx RPC call to the node will take the current provisional State and call State::apply_tx. If the node is a staker, this provisional state is later used to participate in the consensus algorithm (e.g. streamlette).
When a slower node syncs its state with a peer at a higher height, it will continuously call apply_block on its current highest state.