[milindsrivastava1997]

Part 1 of #1046. It covers the plain deletions. The row-reference oracle, temporal::reduce and the devtools corpus helpers are in a follow-up PR.

Why

Production never reaches this code, and keeping it costs time. For example, the memo snapshot had to be re-recorded after unrelated memo changes. Removing it cuts about 4,150 lines.

What

Removed:

• old_docs/. Nothing references it, and docs/ replaces it.
• plan-selection/src/search/costing/demand.rs (#[cfg(test)]). Its tests never compared it against the production cost evaluator.
• executor/src/runtime/batch_execution.rs. Production code never calls it. The one test that used it (aggregate/temporal.rs) now builds its two-node DAG directly.
• logical-asap-optimization/src/pass2/memo/shared_producer_snapshot.rs (#[cfg(test)]). The rule tests in pass2/memo/rules/ cover the same behaviour.
• In physical-asap-optimization: optimize_physical_asap_without_query_time_raw_scans, every_assignment_scans_raw_at_query_time and reads_raw_at_query_time, which have no callers. physical_candidate_count is now private and #[cfg(test)].
• run_stage1_to_stage3_from_pass1, which has no callers. With it gone, the private run_with_initial_memo always received an initial memo, so it now takes DagOfLogicalMemoGroups instead of Option<_>.
• In types: hydra_kind_for and default_hydra_params, plus their tests. Two doc comments that linked to hydra_kind_for now state the rule in prose.
• In types: MeasuredCpu, MeasuredResources, and Measurement, which only MeasuredCpu used. Also the test that compared ModeledCpu against MeasuredCpu.
• docs/develop_docs/offline-sketch-evidence.md, offline-sketch-evidence.schema.json and offline-comparison-evidence.schema.json. These docs described the offline benchmark evidence wire format in terms of MeasuredCpu, Measurement and MeasuredResources. They also named ResourceMeasurements and crates/asap-aware-mapping, neither of which exists anymore. Nothing in this repo produces or reads that format. This deletion is not in Remove dead and test-only code: row-reference oracle, old_docs, test-only demand model, batch bridge, memo snapshot #1046's list.

docs/design_docs/proposals/planner-layering.md no longer names the removed function. The trailing-whitespace pre-commit hook also stripped existing trailing whitespace from 27 lines of that file, so the PR includes those lines too.

How

Deletions only. The one non-deletion code change is the temporal.rs test, which now runs a source → window DAG itself instead of going through evaluate_batch. It still asserts on the same first output row.

As agreed in the planning discussion, this PR leaves the files #1047 rewrites alone: the pass2/memo/* helpers, accuracy::native_target, asap_aware_cse::*, workload_candidates, query_pruning, optimize_physical_asap and select_plan. The one exception is removing the mod shared_producer_snapshot; line in pass2/memo.rs.

Key code interfaces added / modified

Removed public items only. Nothing is added.

// crates/physical-asap-optimization/src/lib.rs
pub fn optimize_physical_asap_without_query_time_raw_scans(..)   // removed
pub fn every_assignment_scans_raw_at_query_time(..) -> bool      // removed
pub fn physical_candidate_count(..)                              // was pub; now private, #[cfg(test)]

// crates/plan-selection/src/search/pipeline.rs (re-exported from search)
pub fn run_stage1_to_stage3_from_pass1<Id>(..)                   // removed

// crates/executor/src/runtime/batch_execution.rs
pub fn evaluate_batch / evaluate_inputs / evaluate_source        // removed (whole module)

// crates/types/src/ir/schema/sketch.rs (re-exported from schema)
pub fn hydra_kind_for(..) / default_hydra_params(..)             // removed

// crates/types/src/workload/resources/
pub struct MeasuredCpu; pub type MeasuredResources;              // removed
pub struct Measurement; pub mod measurement;                     // removed

Major algorithm changes

None.

Before this PR

The workspace carried about 4,150 lines that production never reaches. Every memo change had to re-record the FNV snapshot of the whole memo.

After this PR

The code is gone, and memo changes no longer need a snapshot re-record. Behaviour is unchanged.

Evidence

Not applicable. Behaviour is unchanged; see Verification.

Verification

• Unit tests: cargo test --workspace passes, after rebasing onto 92fab4de (refactor: rename memo group, candidate and computation-target fields (#1047 P2) #1058).
• End-to-end tests: same run.
• Other checks: the pre-commit hooks pass (fmt, check, clippy, test).

No new tests.

Snapshot changes

None.

Architectural decisions

• Test-only public items that other crates' tests/ call can't be put behind #[cfg(test)]. Deciding how to handle them is split out to Decide how to hide test-only public items used across crates #1057.
• Measurement and the evidence docs were removed together with MeasuredCpu, instead of keeping them for a format nothing in the repo uses.

Limitations and follow-up

• PR B: remove the row-reference oracle and temporal::reduce, and deduplicate the devtools corpus helpers.
• Decide how to hide test-only public items used across crates #1057: decide how to hide the remaining test-only public items.

Human review — do not complete with an agent

• The MVP boundary is correct.
• New conceptual layers or public interfaces are necessary.
• The before/after description matches the intended product behavior.
• Human reviewer:
• Decision and rationale:

🤖 Generated with Claude Code