Extract LogEntry/PendingLogEntry traits into a new tlog_entry crate

[lukevalenta]

Third slice of #230. Move LogEntry/PendingLogEntry traits + helpers (and the c2sp.org/tlog-tiles default entry-bundle implementation) out of tlog_tiles into a new internal tlog_entry crate. After this, each spec crate matches its C2SP slug exactly.

Stacked on PR #233 (slice 2: tlog_checkpoint). Diff against #233 is the new content here; once #231/#233 land this PR's base will be retargeted to main.

Why a new crate (and not generic_log_worker)

The natural framing of #230's slice 3 was "move worker abstractions to generic_log_worker". On closer reading, the situation is more nuanced: the traits are a worker ↔ spec interface, with multiple consumers on each side:

• Implementors (spec-level): tlog_tiles (the default tlog-tiles entry-bundle), static_ct_api, bootstrap_mtc_api, soon ietf_mtc_api.
• Consumer (worker-side): generic_log_worker.

generic_log_worker already depends on static_ct_api / bootstrap_mtc_api, so moving the trait definitions there would create a cycle. Putting them in any single spec crate forces every other spec crate to depend on it. A small interface crate (tlog_entry, publish = false) breaks the tangle cleanly.

What moves to tlog_entry

Symbol	Old home	Notes
PendingLogEntry, LogEntry traits	tlog_tiles::entries	The worker / spec interface.
PendingLogEntryBlob	tlog_tiles::entries	DO storage shape.
LookupKey, LOOKUP_KEY_LEN	tlog_tiles::entries	Cache/dedup key.
TileIterator	tlog_tiles::entries	Helper for parsing entry-bundle tiles.
TlogTilesPendingLogEntry, TlogTilesLogEntry	tlog_tiles::entries	Default c2sp.org/tlog-tiles entry-bundle impl. Co-located with the trait it implements.

Crate has publish = false. Deps: length_prefixed, serde, sha2, tlog_checkpoint, tlog_core, tlog_tiles.

What stays in tlog_tiles

Just tile.rs — the c2sp.org/tlog-tiles HTTP wire format itself (Tile, TlogTile, PathElem, TileReader, TileHashReader, TlogTileRecorder, PreloadedTlogTileReader).

Production deps shrink to just tlog_core. length_prefixed, serde, sha2 dropped entirely; base64/url retained only as dev-deps for the CT integration test in lib.rs.

End state of #230

After slices 1–3 ship, each spec crate is aligned with the C2SP slug it implements:

• tlog_core → RFC 6962 Merkle math + draft-ietf-plants-merkle-tree-certs subtree proofs.
• tlog_checkpoint → c2sp.org/tlog-checkpoint signed-note format.
• tlog_tiles → c2sp.org/tlog-tiles HTTP wire format.
• tlog_entry → internal worker/spec interface (publish = false).

Migration

Every consumer of the moved symbols (bootstrap_mtc_api, bootstrap_mtc_worker, ct_worker, generic_log_worker, integration_tests, static_ct_api) now imports from tlog_entry directly.

bootstrap_mtc_worker and ct_worker no longer depend on tlog_tiles at all (they were only pulling it in for the trait definitions).

Stats

25 files changed, +109 / −44 (vs slice-2 head). The entries.rs move is git mv with 87% similarity (the diff is the new module doc-comment header + the tlog_tiles::PathElem import).

Pre-push checks

All four pass:

• cargo clippy --workspace --all-targets -- -Dwarnings -Dclippy::pedantic
• cargo test (33 test result blocks ok, 2 more than slice 2 from tlog_entry's relocated test_parse_tile_entry)
• cargo fmt --all --check
• cargo machete

Closes #230.

[lukevalenta]

/bonk review

[ask-bonk]

Reviewed slice 3 of #230 (the diff against #233's head, since this is stacked).

Summary

Mechanical extraction of LogEntry / PendingLogEntry and friends from tlog_tiles into a new internal tlog_entry crate. The entries.rs → tlog_entry/src/lib.rs move is a git mv with the only content changes being (a) a new module-level doc comment explaining why the crate exists, and (b) use crate::PathElem → use tlog_tiles::PathElem. Confirmed by diffing 9c1c99a:crates/tlog_tiles/src/entries.rs against 724bd29:crates/tlog_entry/src/lib.rs.

Architectural choice

The "why a new crate (and not generic_log_worker)" framing in the PR body is convincing. generic_log_worker already depends on the spec crates that implement these traits, so housing the trait definitions there would create a cycle; pushing them into any single spec crate would force every other spec crate to depend on it. A small publish = false interface crate is the cleanest break, and the boundary aligns with AGENTS.md's rule about keeping worker-specific concerns out of the spec-level crates — tlog_entry is explicitly the worker↔spec interface, not a published spec.

The end state where each remaining spec crate maps 1:1 to its C2SP slug (tlog_core → RFC 6962 + MTC subtree math, tlog_checkpoint → c2sp.org/tlog-checkpoint, tlog_tiles → c2sp.org/tlog-tiles HTTP wire format) is a noticeable improvement over the pre-#230 status quo.

What tlog_tiles becomes

After this slice, tlog_tiles is just tile.rs (HTTP wire format types). Production deps shrink to tlog_core only — length_prefixed, serde, sha2, tlog_checkpoint all dropped, and base64 / serde / serde_json / url retained only as dev-deps for the CT integration test in lib.rs. Spot-checked tile.rs for residual usage of serde / tlog_checkpoint: none. The dep diet is genuine.

Migration consistency

All 11 import-rewrite sites (bootstrap_mtc_api, bootstrap_mtc_worker ×3, ct_worker ×3, generic_log_worker ×4, integration_tests, static_ct_api) consistently switch to tlog_entry::{...}. Where a file needs both moved trait/type symbols and the still-in-tlog_tiles PathElem / tile-reader symbols, the imports are correctly split (e.g. bootstrap_mtc_api/src/lib.rs:29-30, static_ct_api/src/static_ct.rs:124-125, generic_log_worker/src/log_ops.rs:44-45, integration_tests/src/assertions.rs:19-20). No stragglers found.

bootstrap_mtc_worker/Cargo.toml and ct_worker/Cargo.toml correctly drop tlog_tiles entirely — confirmed by grep that neither worker references tlog_tiles:: anywhere after the rewrite.

Dep ordering nit

In bootstrap_mtc_api, bootstrap_mtc_worker, ct_worker, generic_log_worker, and static_ct_api the new line lands as tlog_core / tlog_entry / tlog_checkpoint / tlog_tiles, which isn't strict alphabetical but matches the pre-existing logical order (tlog_core before tlog_checkpoint was already there). integration_tests/Cargo.toml is the one place the slot is alphabetised (tlog_checkpoint / tlog_core / tlog_cosignature / tlog_entry / tlog_tiles / tlog_witness). Worth a follow-up to converge on one convention, but not a blocker for this PR.

Pre-push checks

Re-ran all four locally on 724bd29:

• cargo clippy --workspace --all-targets -- -Dwarnings -Dclippy::pedantic — clean.
• cargo test — 33 test result: ok. blocks, 0 failures (matches the PR body claim exactly; the +2 vs slice 2 is test_parse_tile_entry showing up in the new tlog_entry crate plus its doctest module).
• cargo fmt --all --check — clean.
• cargo machete — clean (no unused deps; the tlog_entry workspace dep is actually used in every crate that declares it).

Verdict

LGTM. Mechanical, well-scoped, and motivated by a real cycle constraint. Once #231/#233 land and this gets rebased onto main, it should merge cleanly.

[ask-bonk]

Posted the review on PR #234. Summary of findings:

• Move is mechanical and clean: entries.rs → tlog_entry/src/lib.rs is a git mv with only a new doc comment + one import rewrite (crate::PathElem → tlog_tiles::PathElem).
• Architectural choice is sound: a publish = false interface crate is the cleanest way to break what would otherwise be a cycle between spec crates and generic_log_worker.
• tlog_tiles slimming is genuine: production deps drop to just tlog_core; base64/serde/url correctly demoted to dev-deps (verified tile.rs doesn't use them).
• Migration consistency: all 11 import-rewrite sites switched correctly; bootstrap_mtc_worker and ct_worker correctly drop tlog_tiles entirely.
• All four pre-push checks pass locally on 724bd29: clippy clean, 33 test result blocks (matches PR claim), fmt clean, machete clean.
• One minor nit: dep ordering convention diverges between integration_tests (alphabetical) and the other 5 crates (logical core/entry/checkpoint/tiles order). Pre-existing convention, not a blocker.

github run