PerryTS · proggeramlug · Jul 25, 2026 · Jul 24, 2026 · Jul 25, 2026 · Jul 25, 2026
diff --git a/changelog.d/6812-builder-fold.md b/changelog.d/6812-builder-fold.md
@@ -0,0 +1 @@
+perf(hir): #6812 — straight-line builder sequences (`const o = {…}; o.k = v; …`) fold into the object literal they spell out before lowering, routing them through the anon-shape literal machinery (shape-cached keys, typed slots, direct stores) instead of N dynamic transition writes. The matrix's worst real-world row (fresh-object builds) improves 7.6× (186× → 25× vs node); conservative guards (`__proto__`, duplicates, accessor/spread literals, self-referencing values) keep the rewrite unobservable — a non-matching shape lowers exactly as before.
diff --git a/changelog.d/6812-clone-eligibility.md b/changelog.d/6812-clone-eligibility.md
@@ -0,0 +1 @@
+perf(codegen): #6812 — widen whole-loop object-write clone eligibility: multiplication in numeric RHS (endpoint-product interval bounds), `let`-declared receiver aliases (reassignment is structurally impossible in the matched region), and the `let i = 0; while (i < N)` spelling of the counted inner loop. Three matrix rows move from 6–9× node to 0.62–0.75× (beating node); the emitter needed no changes — it already finalizes counter slots on the fast edge. Also restores the packed-f64 index-offset matcher's exact Add|Sub guard after a mis-scoped widening was caught pre-merge (multiplied-index sanity added to the validation set).
diff --git a/crates/perry-codegen/src/stmt/loops.rs b/crates/perry-codegen/src/stmt/loops.rs
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		perf(hir): #6812 — straight-line builder sequences (`const o = {…}; o.k = v; …`) fold into the object literal they spell out before lowering, routing them through the anon-shape literal machinery (shape-cached keys, typed slots, direct stores) instead of N dynamic transition writes. The matrix's worst real-world row (fresh-object builds) improves 7.6× (186× → 25× vs node); conservative guards (`__proto__`, duplicates, accessor/spread literals, self-referencing values) keep the rewrite unobservable — a non-matching shape lowers exactly as before.
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		perf(codegen): #6812 — widen whole-loop object-write clone eligibility: multiplication in numeric RHS (endpoint-product interval bounds), `let`-declared receiver aliases (reassignment is structurally impossible in the matched region), and the `let i = 0; while (i < N)` spelling of the counted inner loop. Three matrix rows move from 6–9× node to 0.62–0.75× (beating node); the emitter needed no changes — it already finalizes counter slots on the fast edge. Also restores the packed-f64 index-offset matcher's exact Add\|Sub guard after a mis-scoped widening was caught pre-merge (multiplied-index sanity added to the validation set).