[zxch3n]

Stacked on #1222. The first three commits are #1222 by @serg-and, unchanged; only the last commit (fe33ee1) is new. #1222 does not allow maintainer edits, so the fix could not be pushed to its branch. Merging this supersedes #1222.

Summary

#1222 fixes two costs that grow quadratically with the number of peer ids a document has seen:

• try_shrink_frontiers walked ancestors once per candidate. It now walks once per retained head, and try_vv_to_frontiers returns the maintained dag.frontiers when asked for the current version.
• The movable-list/text/list tracker rebuild (replay_container_ops_from_empty) rebuilt a full version vector for every historical change. It now skips changes that have no op on the container before building one.

The new commit only puts iter_from_replay_base_causally behind #[cfg(test)]. After #1222 it is called only from tests, so cargo clippy -- -Dwarnings failed with dead_code. The doc references now point at try_iter_from_replay_base_causally, the production entry point.

Validation

• Rust port of the perf: speed up concurrent imports with large peer histories #1222 reproduction (5,000 historical peers, 4 concurrent updates, release): the 2nd-4th imports went from ~760 ms to 4-11 ms for Text, Tree and List, and from 2.7-4.6 s to 17-25 ms for MovableList. Map is unchanged. With only the frontier changes, MovableList is still ~450-700 ms, so the container filter is needed too.
• cargo test -p loro-internal -p loro --features loro-internal/test_utils,loro/jsonpath: 1647 passed, 0 failed.
• cargo fmt --all -- --check: clean.
• cargo clippy -p loro-internal --all-features -- -Dwarnings: the dead_code error is gone. The 3 errors left (loro.rs question-mark, query_by_len.rs doc indentation) are also on main with current stable clippy.
• Not run: WASM tests, fuzz.

Notes

• perf: stop the frontier walk at nodes with a cached version vector #1244 cannot be rebased onto this as is. Git merges it without conflicts, but it does not compile: perf: speed up concurrent imports with large peer histories #1222's new walk returns Continue(()) where perf: stop the frontier walk at nodes with a cached version vector #1244 changes the callback to ControlFlow<(), bool>. perf: stop the frontier walk at nodes with a cached version vector #1244's cached-vv stop would also land in the test-only shrink_frontiers_reference, not in the production walk. It needs to be redone in the new walk.

🤖 Generated with Claude Code

[github-actions]

WASM Size Report

• Original size: 3486.50 KB
• Gzipped size: 1168.46 KB
• Brotli size: 818.72 KB

[zxch3n]

Superseded: #1222 was merged directly, and the clippy fix from this PR landed in #1251.