[thekevinbot]

Summary

Stat a path-table row only when the statement reads size, mtime, ctime or content. Stacked on #1317.

Fixes #1318

• Building rows no longer stats anything. A row's stat facts fill lazily.
• The first stat a statement asks for stats that statement's whole row set at once, in parallel. Stats are never done one by one as SQLite steps.
• Content prefetch is unchanged.

Red commit first: build_rows_stats_no_file asserts that building rows stats no file. CI on f51c5b9 failed on exactly that assertion (vtab.rs:713).

Measurements (release build, warm cache)

Query	Base (#1317)	This PR
SELECT path FROM './*', 1M flat files	2.4s	1.6s
Join of two 150k-row tables reading size and content	1.6–2.3s	1.6s

The output is byte-identical to the base in both cases.

An earlier version stat'ed lazily one row at a time. It made the join 7.8s, so it was replaced before push.

glob_speed_one_level_test (#1305) still fails: dirsql takes 1.30s against a 0.38s budget. The walk is the next fix (#1320).

E2E Verification

• Python: uv run python -m pytest tests/e2e/ -q --deselect tests/e2e/glob_speed_one_level_test.py, 40 passed.
  • The deselected test is the known-red speed test this stack is working toward. The receipt records the deselect. Whether a receipt may exclude a known-red test is for the maintainer to decide.
• TypeScript: just e2e-attest-ts (pnpm test:e2e) passed.
• Rust: cargo test --workspace --features cli passed apart from flaky server watcher tests. post_query_returns_rows_over_http and server_drops_the_rows_of_a_directory_moved_out_of_the_root fail the same way on the base branch under load and pass alone.
• By hand:
  • SELECT * returns the stat and content values.
  • SELECT path, size returns correct sizes.
  • SELECT path returns paths only.

Changelog / Migrations

• packages/rust/changelog.d/2026-10-05-path-table-lazy-stat.md
• packages/rust/migrations.d/2026-10-05-path-table-lazy-stat.md (nothing breaks)