[Nic-dorman]

Summary

Repins antd from ant-core 0.10.0 (ant-cli-v0.3.8) to 0.12.0 (ant-cli-v0.3.10, released 2026-10-08), before the antd release that ships byte-range downloads (#313) and the chunk-cache cap (#314). No source changes were needed.

This is an interim repin. A later one will pick up single-flight chunk fetches (WithAutonomi/ant-client#218) once that's released.

What changes

antd/Cargo.toml:

• ant-core: tag ant-cli-v0.3.8 → ant-cli-v0.3.10.
• ant-protocol: =3.0.0 → =4.0.0, which must match ant-core's own copy exactly.
• evmlib (0.10.0) and self_encryption (0.36.0) are unchanged; ant-core 0.12.0 still resolves to them.

antd/Cargo.lock moves four crates (ant-core, ant-protocol 4.0.0, saorsa-core 0.29.0, saorsa-transport 0.38.0) and keeps a single copy of each.

What antd picks up

ant-client recorded no release notes for 0.3.9 or 0.3.10, so this list comes from the commits.

• Cold-download startup (ant-client 96d63f9): early speculative GETs go only to peers that are already connected. Peer discovery gets the full fallback allowance while it runs. A cold download no longer burns its early GET slots on dead addresses.
• Late holders (ant-client 630d646): a holder that discovery finds after the early allowance is spent is still read, rather than waiting for discovery to finish.
• data_download_range now goes through an internal helper that can reuse records the caller already holds. antd's calls behave as before.
• ant-protocol 4.0.0 adds the pointer messages, appended after every existing variant, so older peers reject them cleanly. antd exposes no pointer API.
  • The versioned quote requests (QuoteRequestV2), with their fallback for older nodes, were already in 0.10.0.

The rest of 0.11 and 0.12 is pointers and browser/WASM fixes.

Test evidence

Unit tests: antd passes 113 tests, and fmt and clippy --all-targets -D warnings are clean.

Devnet, 4.5 GiB: a 25-node devnet on ant-node 0.22.0, the release that pairs with ant-core 0.12.0 (the earlier runs used 0.20.0). The daemon is a release build of this branch (build_commit 00b94aa), so it includes #313–#317 and the #314 cache cap at its default of 64. The object is 4,831,850,553 bytes. ALL PASS, and antd logged no ERROR or WARN lines.

• Range check:
  • 38 ranges over REST and gRPC, private and public, are byte-identical to the source file. They include ranges past 2^32 and the edge of the last chunk.
  • The unsatisfiable cases give 416 with bytes */size, and gRPC OUT_OF_RANGE carries the size (the run requires it).
  • Every range fetched exactly the same chunks as the 0.10.0 run did: the per-range overlap, fetched and cache-hit counts match for all 38.
• antd-rust range methods (feat(antd-rust): byte-range stream methods on the REST and gRPC clients #317): all 30 checks pass on both transports, including whole-object reads and RangeNotSatisfiable { size: Some(4831850553) }.

Compared with the same check on 0.10.0 and ant-node 0.20.0 with the cache cap:

	ant-core 0.10.0	ant-core 0.12.0
Whole object, bytes=0-	67.4 MB/s	70.0 MB/s
Peak antd RSS on that read	887 MiB	914 MiB
gRPC stream from 2^32 − 512 MiB to the end (1.07 GB), peak RSS	903 MiB	904 MiB
Private upload, 1,157 chunks	6m42s (12.0 MB/s)	7m34s (10.6 MB/s)
Public upload	2m21s	2m27s

The upload was about 13% slower. Each figure is a single run on one box, and the node version changed as well as the client. Two earlier 0.10.0 runs took 6m42s and 6m28s.

🤖 Generated with Claude Code