`LoadedPrograms` - Githubissues

One global loaded programs cache per validator with two index levels: Program address and effective slot
One local loaded programs cache per transaction batch
Each entry has a deployment slot (making this a bitemporal model)
And an effective slot, which does not refer to a specific block but all future blocks beyond that slot, which are descendants of the deployment slot
Entries can hold the following types of programs:
- LegacyV0: Programs of the legacy loader(-v1) or loader(-v2)
- LegacyV1: Programs of the upgradeable loader(-v3)
- Typed: Programs of loader-v4
- Built-in
- Tombstone: To remember that a program failed verification or was closed
- Unloaded: To remember that a program was verified before and what usage statistics it had
- Loading: To coordinate cooperative loading across transaction batches
Recompilation phase before RBPF-affecting features activate to amortize workload on epoch boundary
Delay visibility: Avoids pipeline stalls as the very next instruction in a transaction could depend on a redeployment
Redeployment cool down: Prevents multiple (un-/re-)deployments of the same program in a slot
Usage statistics: Kept in atomic counters in the program which itself lives in an Arc
Simplify the transaction loader to deduplicate all accounts and not keep duplicate program accounts
Does not care about the dependency graph in between programs, all transitive dependencies need to be specified in a message

solana-labs / solana