mirrors/arti.git/crates/tor-hscrypto/src/pow.rs, branch main

PoW: Consolidate feature flags into a single flag.

2024-11-25T20:49:46Z

This replaces the hs-pow, hs-pow-full, hs-pow-v1, pow-v1, and pow-full features with a single hs-pow-full feature. It's possible that in the future we will want to split different schemes into different features, but we can do that when it comes up. For now, having this as a single flag makes it clearer what's going on, since the previous thing was not actually expressive enough to capture some things we care about (like "at least one pow scheme is enabled" that works in a future-compatible way). This change is not semver breaking since it's a experimental feature. Related: #1751

tor-hscrypto: Fixup broken doc links.

2024-10-09T16:05:21Z

Co-authored-by: Micah Elizabeth Scott

hs-pow: Refactor to prefer stub types rather than cfg sprawl.

2024-10-09T16:05:15Z

In this design, the thin multiplexing layer between PoW types is always available when onion services are in use, but the specific pow schemes (and their dependency libraries) are gated by crate features everywhere. There are now no new cfg() gates. When the pow-v1 scheme is disabled, we can parse `pow-params v1` lines into an empty type (so clients know a PoW scheme exists that might be supported if they were configured differently). We currently don't save the contents of unknown hsdesc items. On the relaycell side, the hs ext module already sets a strong precedent for keeping unrecognized data as a byte vec, and it doesn't provide a good way to signal soft parse errors like unrecognized optional extensions. There, the `v1` type is completely optional, and services lacking a pow scheme suggested by a client would see one of these 'unrecognized' blobs. This isn't necessarily helpful but it fits the rest of the design. Co-authored-by: Micah Elizabeth Scott

tor-hspow: Big refactor, dissolve this crate

2024-10-09T16:04:35Z

My previous strategy here was to try and centralize hspow in one crate, writing it like a self-contained feature. That introduced friction in the data types, prompting the use of simplistic types at the netdoc/cell layers and full-featured types in the optional modules. This changes tactics, dissolving the low-level parts of tor-hspow into tor-hscrypto and the high-level parts into hsclient/hsservice. Full featured types are used everywhere now, but the tradeoff is that compile-time configurability is a lot more pervasive. Anything that knows about PoW types at all needs to be fully configured out. I took this opportunity to try a more complete set of crate features, allowing users to configure individual PoW schemes. Co-authored-by: Micah Elizabeth Scott