nordisk/skade

aegir — crate card

Reverse-engineered (from source — verify, don't trust the name)

• One-line pitch: Ægir — a tiny pure-Rust S3 client (SigV4 over ureq, zero-copy reads). Historically: "coexists with gix where rust-s3 can't" — see the stale-claim note at the bottom; that is no longer true of current gix, and aegir now earns its keep on size and zero-copy alone.
• Does: Minimal S3 client with AWS Signature V4 (HMAC-SHA256) signing and path-style addressing (RustFS/MinIO/real S3). Public: Client (with Client::with_transport), Transport enum, Error/Result. Reads return bytes::Bytes so bodies pass into arrow/parquet without re-copy. Pluggable byte transport: default ureq (rustls, TLS+plain); opt-in uring feature adds an io_uring TCP transport for the plaintext hot path (falls back to ureq for TLS).
• Key deps: ureq (pure-Rust HTTP/rustls), hmac+sha2 (SigV4, RustCrypto), bytes (zero-copy bodies), chrono (SigV4/RFC1123 timestamps), optional io-uring (feature uring; deps only libc/bitflags/cfg-if — no C, no bindgen).
• Shape: lib only, 2 modules (lib.rs ~17K + feature-gated uring.rs ~17K). Small, no intra-workspace deps.
• Observed role in the workspace: Exists specifically for skade-katalog-bench's S3 backend (--features s3); the bridge into iceberg's Storage/StorageFactory is bench/src/s3_storage.rs. Lives as a workspace member under bench/, publish = false. Originally chosen over rust-s3 for the gix/maybe-async reason below — that reason has expired, the crate has not.

Fill in (one line each — only the human knows these)

• Primary consumer(s):
• Why it exists (vs the obvious alternative):
• Hard constraint (airgap / no-C-deps / no-FFI / pure-Rust / etc.):
• Release cadence & independence (rides a parent? own repo someday?):
• Status (production / prototype / experimental / stubbed):
• Must-NOT-change invariant:
• Biggest known gap or TODO:
• Anything a newcomer would get wrong about it:

Stale-claim check (2026-07-31)

The "coexists with gix where rust-s3 can't" line — repeated in this crate's description, in bench/Cargo.toml, and formerly across skade's own manifest and module docs — was TRUE and is now EXPIRED. Root cause, measured:

• gix-protocol ≤ 0.63 depended on maybe-async and its blocking-client feature enabled maybe-async/is_sync (gix-protocol-0.62.0/Cargo.toml:54 "maybe-async/is_sync",).
• Cargo unifies features per crate across the whole graph, so that single edge flipped the macro to sync mode for every crate in the build — including rust-s3, whose async surface is maybe-async-annotated. rust-s3 then had await inside sync fns: ~39 "Result<..> is not a future" errors.
• gix-protocol 0.64.0 dropped maybe-async entirely (0 occurrences in its manifest) and gix 0.86 requires that line, so nothing can enable is_sync.

Verification recipe (re-run it before trusting either the claim or this refutation):

cargo tree --all-features -e features -i maybe-async
#  -> rust-s3 is the ONLY reverse dep, feature "default" (not "is_sync")
cargo tree --all-features -p gix | grep -c maybe-async     # -> 0

Consequence: aegir is no longer required for gix coexistence. It is kept for what it independently is — small (no hyper / tokio-rustls stack), zero-copy (bytes::Bytes straight into parquet), and the io_uring plaintext transport. If a future gix reintroduces maybe-async, the constraint returns exactly as described above.