Brummel
5170b6abd1
Closes Gitea #1. Realises the "P2 follow-up" called out in examples/prelude.ail:9 — removes the surface comparator names `==` / `!=` / `<` / `<=` / `>` / `>=` from the language entirely, routes the LLM-author's `(app eq …)` / `(app compare …)` / `(app ne|lt|le|gt|ge …)` through prelude.Eq / prelude.Ord class- dispatch, ships six named Float-comparison fns (`float_eq`/`float_ne`/`float_lt`/`float_le`/`float_gt`/`float_ge`) so Float keeps comparability without an Eq/Ord instance, and emits primitive instance bodies with `alwaysinline` so the -O0 IR shape stays at one instruction per comparison. Plan-task journal (13 tasks, single atomic iter per Approach A): Task 1 — Bootstrap: 4 new fixtures (eq_user_adt_smoke.ail, eq_float_must_fail.ail, float_compare_smoke.ail, operator_unbound_check.ail) + 5 new E2E/pin tests as RED starting state. All 5 confirmed RED at start: north-star fixed `NoInstance Eq Unit` (preserved by Unit-eq opening line intentionally added in plan-self-review); float_compare unknown `float_eq`; operator-name typechecked (== still polymorphic); must-fail diagnostic lacked `float_eq`; alwaysinline absent from IR. Task 2 — Codegen alwaysinline + intercept arms: introduced `intercept_emit_wants_alwaysinline` allowlist + appended ` alwaysinline` between `)` and `{` of the `define` line in `emit_fn`; added 9 new intercept arms to `try_emit_primitive_instance_body` — `eq__Int`/`eq__Bool`/ `eq__Unit` + the six `float_*` arms. Task 3 — Prelude reshape: `instance Eq Unit` added; Eq Int/Bool bodies become placeholder-`false` (intercept overrides); six `float_*` free fns added; line-9 P2-follow-up comment removed. Lockstep hash re-pins: prelude_module_hash_pin (3abe0d3fa3c11c99 → new), mono_hash_stability six body-hash literals (eq__Int/Bool/Str + compare__Int/Bool/Str all shifted because placeholder body changes the canonical hash). IR-snapshot regen (hello/list/max3/sum/ws_main) rolled forward into this task at orchestrator's pragmatic call — prelude shift forces the snapshots immediately. Task 4 — Fixture migration: 58 .ail fixtures rewritten (`(app == …)` → `(app eq …)`, etc.; Float-typed operand sites to `(app float_eq …)` / `(app float_lt …)`). eq_ord_user_adt.ail:21 inner `==` → `eq` migration with lockstep eq_ord_e2e.rs:134 body-hash re-pin (3c4cf040cb4e8bb2 → new). Two prose snapshots accepted; deps test + 5 hash_pin literals updated as cascading consequences. Task 5 — Test-scaffold migration: all 8 in-source `#[cfg(test)] mod tests` AST-literal sites threaded (desugar.rs:2414, check/lib.rs:5656/6092/6220, codegen/lib.rs sites). 7 obsolete in-source tests deleted (5 eq_typechecks + 2 lower_eq ADT/Fn rejection — coverage moves to E2E eq_user_adt_smoke + eq_float_must_fail). 2 additional letrec tests deleted (single-module check env can't resolve eq/ge without prelude auto-import; covered by workspace E2E). Task 6 — Lit-pattern desugar: build_eq emits `Term::Var { name: "eq" }` instead of `"=="` at desugar.rs:1109; doc-comment rewritten to describe class-dispatch. Task 7 — Dead-machinery deletion sweep (compile-gated): typchecker builtins (install + list comparator entries deleted); codegen builtin_binop_typed (10 comparator arms deleted from synth.rs, table reduced to arithmetic core); lower_eq fn entirely deleted; `==` short-circuit in lower_app deleted; `is_static_callee` `==` clause deleted; `is_arithmetic_or_comparison_op` renamed to `is_arithmetic_op` + caller-update at codegen/lib.rs:2152 + :2529. Dead `poly_a_a_to_bool` helpers also removed. Workspace build green after compile gate — every caller of the deleted surface was migrated by Tasks 4-6. Task 8 — Float-aware NoInstance diagnostic: check/lib.rs:856-880 addendum extended to name `float_eq` / `float_lt` as the explicit alternative for Eq/Ord at Float; eq_float_noinstance.rs:32-44 assertion extended. Task 9 — IR snapshot regen: subsumed by Task 3 (prelude shift forced immediate snapshot regen; deferring to Task 9 would have left the workspace red between tasks). Task 10 — Prose-projection cleanup: 6 comparator arms deleted from binop_info; 3 in-source mod-tests updated/deleted (comparator-infix rendering would be dishonest now that the operators are no longer language identifiers). Task 11 — Contract updates: 5 design files rewritten to the class-dispatch present — float-semantics.md (arithmetic guarantees retained on +/-/*/; comparison guarantees transferred from ==/!=/< to float_eq/float_ne/float_lt/etc.); prelude-classes.md (Eq Unit added to instance list; new paragraph on six float_* fns; Float-no-Eq/Ord clause gains `→ use float_eq` cross-reference); str-abi.md (clause "REMAIN primitive operators" rewritten to describe class-method dispatch); scope-boundaries.md (multiple operator-name and Pattern::Lit-desugar clauses rewritten); authoring-surface.md (==, <= dropped from operator-example list). Task 12 — Initial acceptance gate: workspace 638/0 GREEN; bench/compile_check + cross_lang 0 regressed; bench/check.py flagged 4 regressions on bench_closure_chain (+29% bump_s, +47% rc_s, +29% bump_rss_kb, +48% rc_rss_kb). Orchestrator initially classified as DONE-with-concerns; Boss reclassified as PARTIAL-via-acceptance-#8-fail after independent re-run, extended iter scope to Task 13. Task 13 — Direct icmp intercept arms (Boss-extension): try_emit_primitive_instance_body gains direct-icmp arms for lt__Int / le__Int / gt__Int / ge__Int / ne__Int, bypassing the compare__Int → Ordering → match indirection that allocated one Ordering ctor per call in tight loops. Each new arm inherits `alwaysinline` via the existing allowlist (extended accordingly). New IR pin test `ord_int_intercept_ir_pin.rs` + smoke fixture `ord_int_intercept_smoke.ail` ratify the optimization — opt -O2 -S confirms zero `call @ail_prelude_lt__Int` in optimized IR; the icmp folds directly at every use site. Bool variants (lt__Bool/etc.) deliberately NOT added — no bench/example calls Ord at Bool; spec-extension permits skipping if unreachable at bench level. Family can be extended symmetrically when first Bool-ordered perf workload appears. Bench-gate post-Task-13: bench_closure_chain bump_s -6.05%, rc_s -2.67%, bump_rss_kb -0.77%, rc_rss_kb +0.46% — all four previously-regressed metrics back inside tolerance. Full bench corpus: 36 metrics, 0 regressed, 0 improved beyond tolerance, 36 stable. bench/check.py exit 0 ✓; bench/compile_check.py exit 0 ✓; bench/cross_lang.py exit 0 ✓. Workspace: 640 passed, 0 failed across all binaries (delta vs. milestone start: +5 new E2E/pin tests added in Task 1 + 1 IR pin added in Task 13 − 9 in-source mod tests deleted as obsolete net ≈ −3). The eq_user_adt_smoke fixture's Unit-opening line was the deliberate RED-first device added in planner self-review so the north-star wouldn't accidentally GREEN at start (user-ADT-Eq on Point with hand-written instance was already operable today; the milestone's actual delivery is operator-name death + Eq Unit + Float-named-fns + cleanup of the two-pathy primitive comparator machinery). Net delta: 96 files changed, 1760 insertions, 1101 deletions; 12 net-new files (6 new fixtures, 5 new E2E/pin tests, 1 stats file). main passes-test-count: 640 (was 633 pre-iter, accounting for the deletions). Spec-vs-acceptance addendum: spec §Testing strategy anticipated the bench-gate-regression case with two recovery paths (`alwaysinline` investigation or "spec needs revisiting toward α"). Task 13 is the third path — direct intercept arms for `lt`/etc. that bypass the compare→match path entirely. The spec's contingency clause was thus generous enough to absorb Task 13 without spec revision, but a future iter that hits a class-method primitive where the body shape introduces a similar codegen cost (Ordering allocation, RC tax, deferred-init) should expect a parallel extension. The pattern is "primitive instance whose canonical body indirects through other class methods that allocate" → add a direct-emit intercept arm + alwaysinline + IR-shape pin. Concerns absorbed: - Codegen lower_app / resolve_top_level_fn gained a prelude-fallback lookup so bare monomorphic prelude fns (`float_eq` etc.) resolve from non-prelude modules without explicit `prelude.float_eq` qualifier. Mirrors the typechecker's implicit prelude import — not in plan but necessary infra for the named-fn surface to be callable. - Recon's "≈10 fixtures" estimate was 6× too low — 58 actual. Mechanical migration; no design impact. - Recon caught 4 in-source AST-literal sites the spec missed (lib.rs:5656 `>=` site + three codegen/lib.rs sites); Task 5 covered all. - Recon caught 2 contract files outside the spec's update set that directly contradicted the milestone (str-abi.md + scope-boundaries.md "REMAIN primitive operators" / "Pattern::Lit desugar to ==" clauses); Task 11 covered all 5. Stats file: `bench/orchestrator-stats/2026-05-21-iter-operator-routing-eq-ord.1.json`. closes #1
170 lines
5.7 KiB
Plaintext
170 lines
5.7 KiB
Plaintext
; Latency-distribution bench fixture — explicit-mode variant.
|
|
;
|
|
; Companion to bench_latency_implicit. Same algorithm, with
|
|
; `(borrow T)` / `(own T)` annotations so that under --alloc=rc the
|
|
; codegen emits proper inc/dec instrumentation: the persistent tree
|
|
; cache is borrowed (no inc/dec on pin_root's hot path), the per-op
|
|
; IntList is owned by sum_list (param drop at fn return frees the
|
|
; chain). The drop-iterative annotation on IntList keeps the per-op
|
|
; deallocation O(1)-stack regardless of list length.
|
|
;
|
|
; This is the "RC-fair" arm of the latency bench; together with
|
|
; bench_latency_implicit (the implicit-mode control arm at rc) it
|
|
; tests the hypothesis "RC under explicit-mode has p99
|
|
; per-operation latency within a small constant factor of the
|
|
; median, even under continuous alloc pressure with a large
|
|
; persistent live working set".
|
|
;
|
|
; Workload (must match bench_latency_implicit's parameters exactly
|
|
; for the comparison to be fair):
|
|
; - Live cache: balanced binary tree of depth 19 (524_287 nodes,
|
|
; ~16 MB), borrowed throughout the loop.
|
|
; - Per-op: build a CHUNK_LEN-cell IntList of 0..CHUNK_LEN-1, sum
|
|
; it, drop it. CHUNK_LEN, NUM_OPS, PRINT_K hardcoded to match
|
|
; the implicit variant.
|
|
; - Same stdout shape: one READY (8888), N_PRINT timing markers
|
|
; each containing the per-op sum (validatable, always equal),
|
|
; one DONE (9999).
|
|
|
|
(module bench_latency_explicit
|
|
|
|
(data Tree
|
|
(doc "Balanced binary tree, 32-byte cells. Borrowed across the bench loop; tree-depth recursion at scope close is bounded (depth 19) so no drop-iterative needed.")
|
|
(ctor TLeaf)
|
|
(ctor TNode (con Int) (con Tree) (con Tree)))
|
|
|
|
(data IntList
|
|
(doc "Singly-linked Int list. Per-op chains are CHUNK_LEN long; (drop-iterative) keeps Own-param drop O(1) stack regardless of length.")
|
|
(ctor LNil)
|
|
(ctor LCons (con Int) (con IntList))
|
|
(drop-iterative))
|
|
|
|
; ---------- Live cache: balanced tree of given depth ----------
|
|
|
|
(fn build_tree
|
|
(doc "Build a balanced tree of given depth, every value = 1. Returns owned Tree; main holds it across the loop and the final drop fires at main's scope close.")
|
|
(type
|
|
(fn-type
|
|
(params (con Int))
|
|
(ret (own (con Tree)))))
|
|
(params depth)
|
|
(body
|
|
(if (app eq depth 0)
|
|
(term-ctor Tree TLeaf)
|
|
(term-ctor Tree TNode
|
|
1
|
|
(app build_tree (app - depth 1))
|
|
(app build_tree (app - depth 1))))))
|
|
|
|
(fn pin_root
|
|
(doc "Constant-time tree liveness pin — read root tag, return 1 (TNode) or 0 (TLeaf). Borrows t so the persistent cache is not inc/dec'd on every op.")
|
|
(type
|
|
(fn-type
|
|
(params (borrow (con Tree)))
|
|
(ret (con Int))))
|
|
(params t)
|
|
(body
|
|
(match t
|
|
(case (pat-ctor TLeaf) 0)
|
|
(case (pat-ctor TNode v l r) 1))))
|
|
|
|
; ---------- Per-op work: build/sum an N-cell list ----------
|
|
|
|
(fn cons_n_acc
|
|
(doc "Tail-recursive list builder. Returns owned chain; caller is sum_list which owns and drops it.")
|
|
(type
|
|
(fn-type
|
|
(params (con Int) (own (con IntList)))
|
|
(ret (own (con IntList)))))
|
|
(params n acc)
|
|
(body
|
|
(if (app eq n 0)
|
|
acc
|
|
(tail-app cons_n_acc
|
|
(app - n 1)
|
|
(term-ctor IntList LCons (app - n 1) acc)))))
|
|
|
|
(fn cons_n
|
|
(doc "Build [0,1,...,n-1] :: IntList. Returns owned chain.")
|
|
(type
|
|
(fn-type
|
|
(params (con Int))
|
|
(ret (own (con IntList)))))
|
|
(params n)
|
|
(body
|
|
(app cons_n_acc n (term-ctor IntList LNil))))
|
|
|
|
(fn sum_list_acc
|
|
(doc "Tail-recursive sum. Owns xs; consumes it via the LCons arm's t binder (move-into-tail-call).")
|
|
(type
|
|
(fn-type
|
|
(params (own (con IntList)) (con Int))
|
|
(ret (con Int))))
|
|
(params xs acc)
|
|
(body
|
|
(match xs
|
|
(case (pat-ctor LNil) acc)
|
|
(case (pat-ctor LCons h t)
|
|
(tail-app sum_list_acc t (app + acc h))))))
|
|
|
|
(fn sum_list
|
|
(doc "Sum every element. Owns xs, hands it to sum_list_acc which consumes it.")
|
|
(type
|
|
(fn-type
|
|
(params (own (con IntList)))
|
|
(ret (con Int))))
|
|
(params xs)
|
|
(body
|
|
(app sum_list_acc xs 0)))
|
|
|
|
(fn one_op
|
|
(doc "One bench operation: build+sum a fresh CHUNK_LEN-cell list, pin the tree's root, return their sum so the value chain stays observable. Tree is borrowed; no inc/dec on the hot path against the persistent cache.")
|
|
(type
|
|
(fn-type
|
|
(params (con Int) (borrow (con Tree)))
|
|
(ret (con Int))))
|
|
(params chunk_len t)
|
|
(body
|
|
(app + (app sum_list (app cons_n chunk_len)) (app pin_root t))))
|
|
|
|
; ---------- Bench loop ----------
|
|
|
|
(fn loop
|
|
(doc "Tail-recursive bench loop. Tree is borrowed across all iterations.")
|
|
(type
|
|
(fn-type
|
|
(params (con Int) (con Int) (con Int) (con Int) (borrow (con Tree)))
|
|
(ret (con Unit))
|
|
(effects IO)))
|
|
(params remaining print_countdown chunk_len print_k t)
|
|
(body
|
|
(if (app eq remaining 0)
|
|
(app print 9999)
|
|
(if (app eq print_countdown 0)
|
|
(seq
|
|
(app print (app one_op chunk_len t))
|
|
(tail-app loop
|
|
(app - remaining 1)
|
|
(app - print_k 1)
|
|
chunk_len
|
|
print_k
|
|
t))
|
|
(let _v (app one_op chunk_len t)
|
|
(tail-app loop
|
|
(app - remaining 1)
|
|
(app - print_countdown 1)
|
|
chunk_len
|
|
print_k
|
|
t))))))
|
|
|
|
(fn main
|
|
(doc "Top-level: build tree, signal READY (8888), run loop, signal DONE (9999 emitted by loop).")
|
|
(type (fn-type (params) (ret (con Unit)) (effects IO)))
|
|
(params)
|
|
(body
|
|
(let t (app build_tree 19)
|
|
(let _root (app pin_root t)
|
|
(seq
|
|
(app print 8888)
|
|
(app loop 20000 0 500 20 t)))))))
|