Replace constant hashmaps with const fn

[vmx]

Description

Currently there are some constants defined as hashmaps in

rust-fil-proofs/filecoin-proofs/src/constants.rs

Lines 51 to 120 in 4ff161d

	pub static ref POREP_MINIMUM_CHALLENGES: RwLock<HashMap<u64, u64>> = RwLock::new(
	[
	(SECTOR_SIZE_2_KIB, 2),
	(SECTOR_SIZE_4_KIB, 2),
	(SECTOR_SIZE_16_KIB, 2),
	(SECTOR_SIZE_32_KIB, 2),
	(SECTOR_SIZE_8_MIB, 2),
	(SECTOR_SIZE_16_MIB, 2),
	(SECTOR_SIZE_512_MIB, 2),
	(SECTOR_SIZE_1_GIB, 2),
	(SECTOR_SIZE_32_GIB, 176),
	(SECTOR_SIZE_64_GIB, 176),
	]
	.iter()
	.copied()
	.collect()
	);
	pub static ref POREP_PARTITIONS: RwLock<HashMap<u64, u8>> = RwLock::new(
	[
	(SECTOR_SIZE_2_KIB, 1),
	(SECTOR_SIZE_4_KIB, 1),
	(SECTOR_SIZE_16_KIB, 1),
	(SECTOR_SIZE_32_KIB, 1),
	(SECTOR_SIZE_8_MIB, 1),
	(SECTOR_SIZE_16_MIB, 1),
	(SECTOR_SIZE_512_MIB, 1),
	(SECTOR_SIZE_1_GIB, 1),
	(SECTOR_SIZE_32_GIB, 10),
	(SECTOR_SIZE_64_GIB, 10),
	]
	.iter()
	.copied()
	.collect()
	);
	pub static ref LAYERS: RwLock<HashMap<u64, usize>> = RwLock::new(
	[
	(SECTOR_SIZE_2_KIB, 2),
	(SECTOR_SIZE_4_KIB, 2),
	(SECTOR_SIZE_16_KIB, 2),
	(SECTOR_SIZE_32_KIB, 2),
	(SECTOR_SIZE_8_MIB, 2),
	(SECTOR_SIZE_16_MIB, 2),
	(SECTOR_SIZE_512_MIB, 2),
	(SECTOR_SIZE_1_GIB, 2),
	(SECTOR_SIZE_32_GIB, 11),
	(SECTOR_SIZE_64_GIB, 11),
	]
	.iter()
	.copied()
	.collect()
	);
	// These numbers must match those used for Window PoSt scheduling in the miner actor.
	// Please coordinate changes with actor code.
	// https://github.com/filecoin-project/specs-actors/blob/master/actors/abi/sector.go
	pub static ref WINDOW_POST_SECTOR_COUNT: RwLock<HashMap<u64, usize>> = RwLock::new(
	[
	(SECTOR_SIZE_2_KIB, 2),
	(SECTOR_SIZE_4_KIB, 2),
	(SECTOR_SIZE_16_KIB, 2),
	(SECTOR_SIZE_32_KIB, 2),
	(SECTOR_SIZE_8_MIB, 2),
	(SECTOR_SIZE_16_MIB, 2),
	(SECTOR_SIZE_512_MIB, 2),
	(SECTOR_SIZE_1_GIB, 2),
	(SECTOR_SIZE_32_GIB, 2349), // this gives 125,279,217 constraints, fitting in a single partition
	(SECTOR_SIZE_64_GIB, 2300), // this gives 129,887,900 constraints, fitting in a single partition
	]
	.iter()
	.copied()
	.collect()

. They are wrapped in a RwLock which is only needed for prodbench, which I even don't know whether anyone is still using/if it still works. So wither prodbench can be removed and if not, I think it doesn't really need the feature of being able to add custom constants.

This means that those hashmaps could be replaces with a constant functions. E.g. POREP_PARTITIONS could be replaces with something like:

pub const fn porep_partitions(sector_size: u64) -> usize {
       match sector_size {
           SECTOR_SIZE_2_KIB => 1,
           SECTOR_SIZE_4_KIB => 1,
            …
           _ => panic!("invalid sector size"),
       }
}

The advantage is that we don't need a lazy_static, but also that if you pass in an unsupported size, it will panic at compile-time, not at runtime.

This change would also remove a lot of boilerplate code. If you search for POREP_PARTITIONS, you'll find many occurrences of:

*POREP_PARTITIONS
    .read()
    .expect("POREP_PARTITIONS poisoned")
    .get(&SECTOR_SIZE_2_KIB)
    .expect("unknown sector size")

which could be replaced by porep_partitions(SECTOR_SIZE_2_KIB).

Acceptance criteria

The hashmaps mentioned above are replaced by const functions.

Risks + pitfalls

Those are part of the public API, so crates like filecoin-proofs-api would also need to be adapted.

Where to begin

Change one of those hashmaps to a const fn and get things compiled.

[DrPeterVanNostrand]

It would be really cool (imho) if we made all proofs' code generic over <const SECTOR_NODES: usize>, which would allow Groth16's usage of the const fn some_setup_param<const SECTOR_NODES: usize>() pattern and would result in many simplifications around proof setup/configuration.