linq: table group_by fusion — TableAdapter rides plan_group_by_core

[borisbat]

What

each_kv(tab) / keys(tab) / values(tab) chains with group_by declined at the can_group_by_source gate (TableAdapter inherited the base false) and cascaded to tier-2: materialize a kv array, re-enter the array lane. That cascade was the entire groupby_* gap in the m7 column.

The group_by splice pattern is fully adapter-generic (can_group_by_source → build_group_by_adapter → plan_group_by_core), so enabling tables is two overrides on TableAdapter, mirroring DecsAdapter:

• can_group_by() → true
• build_group_by_adapter() → fresh TableAdapter. The upstream_join arm returns null: join |> group_by over a table lead stays on tier-2 (named deferred edge in LINQ_TO_TABLE.md; the stage-5 terminator-path join is untouched).

plan_group_by_core hands the bucket-fill body to wrap_source_loop, so the kv usage-pruner sees the whole accumulation (key expr + reducer updates + upstream where/select segments) — a group key over kv.value.brand walks values(tab) alone. The non-copyable-values gate and the elemType const-scrub are inherited from adapter construction.

Numbers

benchmarks/sql m7 (table) groupby_* family, ns/op (results.md re-swept in this PR):

	before	after
INTERP (count / family range)	163 / 144–201	31 / 30–50
JIT (count / family range)	43.5 / 44–73	8.4 / 8.4–11

~5× in both tiers — the JIT side compounds with #3100's inline slot walk. join_groupby_* unchanged (cascade, by design); INTERP matrix flat elsewhere.

Tests

8 new sub-tests in tests/linq/test_linq_table_source.das (67/67 INTERP + JIT, AOT-compiled in test_aot): kv/keys/values lanes, upstream where+select segments, having + trailing where + order, count terminator, empty table, fused-vs-tier-2 agreement.

Gates

• full INTERP 11001 / 0 failed, full AOT 10313 / 0 failed
• JIT: linq 1971, jit_tests 295 — pass
• CI lint 0 issues; formatter clean
• docs: Sphinx latex + html -W clean, pdflatex pass 1 clean (linq_fold_patterns.rst table-source row updated + new table group_by pattern row; LINQ_TO_TABLE.md stage 7 status + findings + deferred edge)

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Enables group_by fusion for table-backed LINQ sources (each_kv(tab), keys(tab), values(tab)) by letting TableAdapter participate in the existing adapter-generic plan_group_by_core path, avoiding the tier-2 “materialize kv array then regroup” cascade.

Changes:

• Enable table-source group_by fusion via TableAdapter capability + adapter construction hook.
• Add new table-source group_by tests covering kv/keys/values lanes and composed segments (having/where/order/count/empty).
• Update docs and benchmark notes/results to reflect the new fused behavior and performance deltas.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
daslib/linq_fold_table.das	Turns on group_by support for table sources and wires build_group_by_adapter to reuse plan_group_by_core with a fresh TableAdapter (declines upstream-join shape to keep the documented deferred edge).
tests/linq/test_linq_table_source.das	Adds a new test_table_group_by suite validating fused behavior across lanes and common operator compositions.
doc/source/reference/linq_fold_patterns.rst	Documents the new table-lead group_by fusion pattern and its limits (e.g., join→group_by over table lead still cascades).
benchmarks/sql/results.md	Updates benchmark narrative and measured numbers to reflect the new fused table groupby_* behavior.
benchmarks/sql/LINQ_TO_TABLE.md	Advances table-fold “stage” status and documents the new group_by fusion plus remaining deferred edges.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.