[zzylol]

Part 2 of 2 for #1163; closes #1163. Part 1 (#1167) renamed the serialized names.

Why

#1163 decided that "pane", our own word, becomes "tumbling window", the standard stream-processing term, everywhere. After part 1, the operator and its fields carry the new names, but helper types, functions, files, test names, comments and docs still say "pane". This PR finishes the rename so the code reads one way, before #1160 adds the raw-data counterparts (RawDataTumblingWindowChain, RawDataMerge).

What

Every remaining "pane" name moves to the tumbling-window wording, with no behavior change:

Kind	Before	After
types	PaneWidthDomain, PaneCandidate, PaneLayout, PaneCoverageError, PaneChainExpansion(Error), ExpandedPaneChain, SharedPaneChain, PaneWidthFrontier	TumblingWindowWidthDomain, TumblingWindowCandidate, TumblingWindowLayout, TumblingWindowCoverageError, TumblingWindowChainExpansion(Error), ExpandedTumblingWindowChain, SharedTumblingWindowChain, TumblingWindowWidthFrontier
functions	expand_pane_chains(_in_workload(_before)), kept_pane_chains, pane_candidates, panes_of, validate_pane_coverage, test_support::pane	expand_tumbling_window_chains(…), kept_tumbling_window_chains, tumbling_window_candidates, tumbling_windows_of, validate_tumbling_window_coverage, test_support::tumbling_window
fields	PlanningRoundReport.pane_width_domain, PlanningReport.pane_widths_pending, ingestion_pane_chains, pane_candidates, panes	tumbling_window_width_domain, tumbling_window_widths_pending, ingestion_tumbling_window_chains, tumbling_window_candidates, tumbling_windows
modules and files	asap_types::ir::operator::pane_chain; tests pane_chain.rs, kll_pane_execution.rs, tumbling_pane_merge.rs, stored_pane_memory_limit.rs; tools/benchmark_pane_search.py; docs/develop_docs/pane-search-compression-evaluation.md	tumbling_window_chain; tumbling_window_chain.rs, kll_tumbling_window_execution.rs, tumbling_window_merge.rs, stored_tumbling_window_memory_limit.rs; tools/benchmark_tumbling_window_search.py; docs/develop_docs/tumbling-window-search-compression-evaluation.md
text	"pane(s)", "pane chain", "tumbling pane" in comments, docs, test names, labels and messages	"tumbling window(s)", "tumbling-window chain", "tumbling window"

Identifiers always say tumbling_window, never a bare window, which would collide with a query's own window. A few names became clearer instead: pane_window_ms → splittable_window_ms, has_pane_window → has_splittable_window, and test constants PANES / PANE_MS / PER_PANE → TUMBLING_WINDOW_COUNT / TUMBLING_WINDOW_WIDTH_MS / ROWS_PER_TUMBLING_WINDOW.

Kept on purpose:

• The deployed wire contract. The code calls these the deployed wire contract, so only the Rust names change; tests check the wire strings.
  • TumblingWindowLayout reads and writes pane_width_ms / pane_origin_ms. Part 1 had switched the first to tumbling_window_width_ms; this PR goes back to the contract name and keeps the new one as an alias.
  • WindowEdgeCoverage::TumblingWindowAligned is "pane_aligned".
  • TumblingWindowCoverageError is ZeroPaneWidth, UnknownPaneOrigin and pane_phase_ms.
  • WindowEdgeCompatibility is "RequiresAlignedPanePhaseOrExactBoundaryResidual".
• Recorded artifacts: paths, plan ids and archived results in development docs, the viewer's recorded example labels, and part 1's compatibility code.

How

A script renames tokens per language (Rust, JS, Python, Markdown and HTML):

• identifier rules for code, including JS template-literal expressions, regex literals and Python f-string fields;
• English rules for comments, docs and strings that contain spaces.

The result is reviewed by hand:

• doubled words fixed ("tumbling-tumbling", "a tumbling window's window");
• hyphenation of compound modifiers;
• sentences that had contrasted panes with tumbling windows rewritten;
• the viewer's labels kept readable.

Checked-in JSON is data, so the script did not rewrite it. Two exceptions:

• the 17 snapshot files whose display and cost labels the code now writes differently;
• one viewer example's three old report keys.

Key code interfaces added / modified

The renames in the table above; no signature or behavior changes. In tools/dag-viewer/stages.js, upgradeTumblingWindowNames now also renames pane_width_domain and pane_widths_pending in documents written before #1163.

The viewer's internal data fields are now camelCase (tumblingWindowOf, tumblingWindowIndex, tumblingWindows; node kind and CSS class tumblingWindow).

Usage

Planning is unchanged; names and text are what change:

cargo run -p asap-devtools --bin stage_pipeline -- --example planner-layering-1 --out run.json

The document's planning report reads "tumbling_window_width_domain" and "tumbling_window_widths_pending". Plan labels read Kll ×5 tumbling windows, and cost details read build … in 6 tumbling windows of 10000 ms …. A document written before #1163 still opens in the viewer with the same plan.

Major algorithm changes

None.

Snapshot changes

The 17 corpus and sketch-plan snapshots that contain chains change only in their labels: ×N panes → ×N tumbling windows, N × 1m panes → N × 1m tumbling windows, and in N panes of → in N tumbling windows of. No plan, cost or status changes.

The third commit fixes a problem part 1 left in 22 snapshots:

• What broke: part 1 renamed pane_retention_ms → tumbling_window_retention_ms in place, but snapshots store keys sorted, so the renamed key belongs after proven_optimal. The nightly corpus-snapshots job on main would fail.
• Fix: the 22 snapshots are regenerated with UPDATE_SNAPSHOTS=1. They equal the checked-in ones as JSON (22 of 22); only the key order changes.

Before this PR

The tumbling-window chain was SummaryTumblingWindowChain, but its helpers, files and docs said "pane".

After this PR

The code, tests, docs and viewer say "tumbling window" throughout. The only remaining "pane" names are the deployed wire strings above, part 1's compatibility aliases and upgrade, and recorded artifacts.

Other notes

• Output-only strings without readers: population_scope complete_evaluation_including_merged_panes (HLL and UnivMon estimator params) and the fuzz witness NewestPanes are renamed as well. Neither is read back from older files.
• Unrelated test target broken on main: asap-frontend-metricsql's test target doesn't compile on main. refactor: one PromQL entry point, schema resolution as a function, dead frontend code (#1096) #1121 changed lower_promql_query_workload's signature after fix(frontend-metricsql): lower vector negation as PromQL does (#1096) #1120 added a call to it. This PR doesn't touch it.

Verification

• cargo fmt --all -- --check and cargo check --workspace --all-targets: clean, except asap-frontend-metricsql's test target, which is broken on main (see above).
• cargo test --workspace --no-fail-fast, without asap-frontend-metricsql: all pass, including the wire-contract tests (window_edge_names_preserve_wire_values, window_edge_compatibility_names_preserve_wire_values).
• Viewer Node tests (stages, memo, enumeration): 44 pass, including a new check that expanded tumbling-window nodes carry the CSS class the viewer styles.
• Viewer Python tests (python3 -m unittest test_viewer test_server): 117 pass.
• Release corpus_snapshots with --include-ignored (the nightly job, not part of PR CI): the first run failed only on the key order described above. A rerun on the rebased head is in progress.
• Rebased onto main at c83aa19. One conflict: main added #[ignore] to a test this PR renames.
• An independent review found a CSS-class regression in the viewer, a historical doc and archived paths/ids rewritten by the script, 17 stale snapshots and wording problems. All are fixed in the second commit.

🤖 Generated with Claude Code