From 947f665896b917df015888c5efcd036dc4a1acc6 Mon Sep 17 00:00:00 2001
From: Chris Tate <chris@ctate.dev>
Date: Tue, 13 Oct 2026 18:45:47 -0500
Subject: [PATCH] Store boolean arrays as one byte per element

---
 docs/content/docs/how-it-works.mdx            |   2 +-
 examples/native-object/README.md              |   2 +-
 packages/cli/test/native-link-info.test.ts    |   4 +-
 .../backend/llvm/dense-array-access.test.ts   |  24 ++-
 .../src/backend/llvm/dense-array-access.ts    |  29 +--
 .../src/backend/llvm/local-array-reads.ts     |  12 +-
 packages/compiler/src/backend/runtime-abi.ts  |   4 +-
 .../src/backend/runtime-pack-targets.test.ts  |   2 +-
 .../compiler/src/backend/runtime-pack.test.ts |   2 +-
 .../test/native-codegen-integration.test.ts   |   4 +-
 .../corpus/5080-boolean-array-storage.ts.json |   6 +
 .../5081-boolean-array-dynamic-views.ts.json  |   6 +
 .../runtime-darwin-arm64/scripts/build.mjs    |   2 +-
 .../runtime-pack-common/scripts/build.mjs     |   2 +-
 packages/runtime/src/scr_array.c              | 158 +++++++++------
 packages/runtime/src/scr_console.c            |   2 +-
 packages/runtime/src/scr_lib.c                |   2 +-
 packages/runtime/src/scr_map.c                |   2 +-
 packages/runtime/src/scr_runtime.h            |  26 ++-
 packages/runtime/test/test_array.c            | 102 ++++++++++
 tests/corpus/5080-boolean-array-storage.ts    | 183 ++++++++++++++++++
 .../5081-boolean-array-dynamic-views.ts       |  23 +++
 .../self-hosting-native-toolchain.test.ts     |   2 +-
 23 files changed, 503 insertions(+), 98 deletions(-)
 create mode 100644 packages/compiler/test/ts7/baselines/order-parity/tests/corpus/5080-boolean-array-storage.ts.json
 create mode 100644 packages/compiler/test/ts7/baselines/order-parity/tests/corpus/5081-boolean-array-dynamic-views.ts.json
 create mode 100644 tests/corpus/5080-boolean-array-storage.ts
 create mode 100644 tests/corpus/5081-boolean-array-dynamic-views.ts

diff --git a/docs/content/docs/how-it-works.mdx b/docs/content/docs/how-it-works.mdx
index 3d1bb011b..d93559b0e 100644
--- a/docs/content/docs/how-it-works.mdx
+++ b/docs/content/docs/how-it-works.mdx
@@ -70,7 +70,7 @@ Without `-o`, artifacts are written to `.scriptc/` next to the input. Different
 
 ### External linking
 
-Program objects define `main` and leave selected `scr_*` runtime functions undefined. A strong reference to `scr_runtime_abi_v17` checks runtime compatibility at link time.
+Program objects define `main` and leave selected `scr_*` runtime functions undefined. A strong reference to `scr_runtime_abi_v18` checks runtime compatibility at link time.
 
 `--print=native-link-info` emits the program object and a JSON recipe containing the runtime pack and link order for external builds. This object ABI is experimental and requires an exact runtime version. See [Native Program Objects](/docs/native-objects) for examples.
 
diff --git a/examples/native-object/README.md b/examples/native-object/README.md
index 1e7124f85..d79619e6e 100644
--- a/examples/native-object/README.md
+++ b/examples/native-object/README.md
@@ -15,4 +15,4 @@ $ ./app-ld
 
 `cc` uses the C compiler as a linker driver. `ld` links the same precompiled runtime objects directly with the selected Apple SDK. The object defines `main`; it is a complete program object, not a library to load into another process. Use `scriptc build --lib --profile ...` for a host-callable static library.
 
-The external object ABI is experimental. Always consume the runtime pack at the exact `runtime_pack.version` reported by the same scriptc installation. The object requires `scr_runtime_abi_v17`, so a mismatched runtime fails during the link instead of starting with an incompatible ABI.
+The external object ABI is experimental. Always consume the runtime pack at the exact `runtime_pack.version` reported by the same scriptc installation. The object requires `scr_runtime_abi_v18`, so a mismatched runtime fails during the link instead of starting with an incompatible ABI.
diff --git a/packages/cli/test/native-link-info.test.ts b/packages/cli/test/native-link-info.test.ts
index 3d79e2806..6dde7fbd3 100644
--- a/packages/cli/test/native-link-info.test.ts
+++ b/packages/cli/test/native-link-info.test.ts
@@ -72,7 +72,7 @@ describe.runIf(supported)("macOS arm64 native link info", () => {
         minimum_os: "14.0",
       },
       program: { object, entry_symbol: "main" },
-      runtime_abi: { version: 17, marker: "scr_runtime_abi_v17" },
+      runtime_abi: { version: 18, marker: "scr_runtime_abi_v18" },
       runtime_pack: {
         kind: "precompiled",
         package: "@scriptc/runtime-darwin-arm64",
@@ -165,6 +165,6 @@ describe.runIf(supported)("macOS arm64 native link info", () => {
       () => null,
       (failure: { stderr?: string }) => failure,
     );
-    expect(error?.stderr).toContain("scr_runtime_abi_v17");
+    expect(error?.stderr).toContain("scr_runtime_abi_v18");
   });
 });
diff --git a/packages/compiler/src/backend/llvm/dense-array-access.test.ts b/packages/compiler/src/backend/llvm/dense-array-access.test.ts
index ce1e30640..dc23ceae2 100644
--- a/packages/compiler/src/backend/llvm/dense-array-access.test.ts
+++ b/packages/compiler/src/backend/llvm/dense-array-access.test.ts
@@ -77,6 +77,22 @@ test.each([32, 64] as const)("strict scalar reads check dense storage inline (%i
   if (bits === 32) expect(fn).toContain("trunc i64");
 });
 
+test.each([32, 64] as const)("boolean reads load one byte per element (%i-bit)", (bits) => {
+  const read: IrExpr = { kind: "arrayGet", arr: ref("b", flags), index, type: BOOL, loc };
+  const llvm = emitLlvmModule(module("flag", BOOL, [{ kind: "return", value: read, loc }]), {
+    pointerBits: bits,
+  });
+  const fn = body(llvm, "flag");
+  const size = bits === 32 ? "i32" : "i64";
+  expect(fn).toMatch(
+    new RegExp(
+      `getelementptr inbounds i8, ptr %t\\d+, ${size} %t\\d+\\n  %t(\\d+) = load i8, ptr .*\\n  %t\\d+ = icmp ne i8 %t\\1, 0`,
+    ),
+  );
+  expect(fn).not.toMatch(/getelementptr inbounds i64/);
+  expect(fn).toMatch(/call i1 @scr_arr_get_bool\(.*\) cold/);
+});
+
 test("reference reads retain inline and keep a cold +1 runtime fallback", () => {
   const read: IrExpr = { kind: "arrayGet", arr: ref("s", words), index, type: STRING, loc };
   // An element removal in the body keeps the read from being lent as a
@@ -114,7 +130,10 @@ test("element writes store inline, extend the length, and release replaced refer
   );
   const fn = body(llvm, "store");
   expect(fn).toMatch(/store double 0x4004000000000000, ptr /);
-  expect(fn).toMatch(/zext i1 .* to i64/);
+  // Boolean elements are one byte each.
+  expect(fn).toMatch(
+    /getelementptr inbounds i8, ptr .*\n  %t\d+ = zext i1 .* to i8\n  store i8 %t\d+, ptr /,
+  );
   expect(fn).toMatch(/store i8 1, ptr /);
   expect(fn).toMatch(/icmp uge i64 .*/);
   expect(fn).toContain("call void @scr_str_release(");
@@ -143,7 +162,8 @@ test.each([32, 64] as const)(
     const size = bits === 32 ? "i32" : "i64";
     expect(fn.match(new RegExp(`icmp ult ${size} `, "g"))).toHaveLength(3);
     expect(fn).toMatch(/store double 0x4004000000000000, ptr /);
-    expect(fn).toMatch(/zext i1 .* to i64/);
+    expect(fn).toMatch(/zext i1 .* to i8/);
+    expect(fn.match(/getelementptr inbounds i8, ptr /g)).toHaveLength(4);
     expect(fn.match(/store i8 1, ptr /g)).toHaveLength(3);
     expect(fn.match(new RegExp(`add nuw ${size} .*, 1$`, "gm"))).toHaveLength(3);
     // Appends never release: the slot past the length holds no value.
diff --git a/packages/compiler/src/backend/llvm/dense-array-access.ts b/packages/compiler/src/backend/llvm/dense-array-access.ts
index f8613ec2d..83163d99f 100644
--- a/packages/compiler/src/backend/llvm/dense-array-access.ts
+++ b/packages/compiler/src/backend/llvm/dense-array-access.ts
@@ -189,14 +189,21 @@ function statePointer(
   return statePtr;
 }
 
+/** The dense slot type: boolean arrays store one byte per element, every
+ * other element kind an 8-byte slot. */
+function slotType(acc: ArrayAccess): "i8" | "i64" {
+  return acc === "bool" ? "i8" : "i64";
+}
+
 function valuePointer(
   host: LlvmEmitterContext,
   guard: Pick<DenseGuard, "offset" | "data">,
+  acc: ArrayAccess,
 ): string {
   const B = host.B;
   const valuePtr = B.tmp();
   B.line(
-    `${valuePtr} = getelementptr inbounds i64, ptr ${guard.data}, ${host.sizeType} ${guard.offset}`,
+    `${valuePtr} = getelementptr inbounds ${slotType(acc)}, ptr ${guard.data}, ${host.sizeType} ${guard.offset}`,
   );
   host.markMemoryPointer(valuePtr, "array:elements");
   return valuePtr;
@@ -244,7 +251,7 @@ export function emitDenseArrayGet(
   B.line(`${present} = icmp eq i8 ${state}, ${SCR_ARR_VALUE}`);
   B.condBr(present, valueLabel, slow);
   B.startBlock(valueLabel);
-  const ptr = valuePointer(host, guard);
+  const ptr = valuePointer(host, guard, acc);
   let fast: string;
   let fastLabel = valueLabel;
   if (acc === "f64") {
@@ -253,8 +260,8 @@ export function emitDenseArrayGet(
   } else if (acc === "bool") {
     const raw = B.tmp();
     fast = B.tmp();
-    B.line(`${raw} = load i64, ptr ${ptr}${host.fieldAliasAttachment(ptr)}`);
-    B.line(`${fast} = icmp ne i64 ${raw}, 0`);
+    B.line(`${raw} = load i8, ptr ${ptr}${host.fieldAliasAttachment(ptr)}`);
+    B.line(`${fast} = icmp ne i8 ${raw}, 0`);
   } else {
     const raw = B.tmp(),
       nonnull = B.tmp();
@@ -373,14 +380,14 @@ export function emitDenseArraySet(
       B.line(`store double ${value}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
     } else if (acc === "bool") {
       const bits = B.tmp();
-      B.line(`${bits} = zext i1 ${value} to i64`);
-      B.line(`store i64 ${bits}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
+      B.line(`${bits} = zext i1 ${value} to i8`);
+      B.line(`store i8 ${bits}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
     } else {
       B.line(`store ptr ${value}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
     }
   };
   const statePtr = statePointer(host, guard);
-  const valuePtr = valuePointer(host, guard);
+  const valuePtr = valuePointer(host, guard, acc);
   let oldState: string | null = null;
   let old: string | null = null;
   if (acc === "ref") {
@@ -444,7 +451,7 @@ export function emitDenseArraySet(
   // Below the bound the slot holds a present value: replace it, leaving
   // the state alone, and release a replaced reference after the store.
   B.startBlock(filledLabel);
-  const filledValuePtr = valuePointer(host, guard);
+  const filledValuePtr = valuePointer(host, guard, acc);
   let filledOld: string | null = null;
   if (acc === "ref") {
     filledOld = B.tmp();
@@ -504,13 +511,13 @@ export function emitDenseArrayPush(
   B.line(`${data} = load ptr, ptr ${dataPtr}${host.fieldAliasAttachment(dataPtr)}`);
   B.line(`${states} = load ptr, ptr ${statesPtr}${host.fieldAliasAttachment(statesPtr)}`);
   const statePtr = statePointer(host, { offset: len, states });
-  const valuePtr = valuePointer(host, { offset: len, data });
+  const valuePtr = valuePointer(host, { offset: len, data }, acc);
   if (acc === "f64") {
     B.line(`store double ${value}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
   } else if (acc === "bool") {
     const bits = B.tmp();
-    B.line(`${bits} = zext i1 ${value} to i64`);
-    B.line(`store i64 ${bits}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
+    B.line(`${bits} = zext i1 ${value} to i8`);
+    B.line(`store i8 ${bits}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
   } else {
     B.line(`store ptr ${value}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
   }
diff --git a/packages/compiler/src/backend/llvm/local-array-reads.ts b/packages/compiler/src/backend/llvm/local-array-reads.ts
index ef3f67bf3..db56b273d 100644
--- a/packages/compiler/src/backend/llvm/local-array-reads.ts
+++ b/packages/compiler/src/backend/llvm/local-array-reads.ts
@@ -466,7 +466,7 @@ function emitDenseReferenceArrayRead(
   no: string,
   slow: string,
   join: string,
-  loadType: "ptr" | "i64" = "ptr",
+  loadType: "ptr" | "i64" | "i8" = "ptr",
 ): void {
   const B = host.B;
   const range = B.newLabel("local.array.range"),
@@ -552,7 +552,11 @@ function emitDenseReferenceArrayRead(
   );
   host.markMemoryPointer(dataPtr, "array:header");
   B.line(`${data} = load ptr, ptr ${dataPtr}${host.fieldAliasAttachment(dataPtr)}`);
-  B.line(`${valuePtr} = getelementptr inbounds i64, ptr ${data}, ${host.sizeType} ${offset}`);
+  // Boolean arrays store one byte per element; every other kind an 8-byte slot.
+  const slotType = loadType === "i8" ? "i8" : "i64";
+  B.line(
+    `${valuePtr} = getelementptr inbounds ${slotType}, ptr ${data}, ${host.sizeType} ${offset}`,
+  );
   host.markMemoryPointer(valuePtr, "array:elements");
   B.line(`${raw} = load ${loadType}, ptr ${valuePtr}${host.fieldAliasAttachment(valuePtr)}`);
   present(raw);
@@ -663,7 +667,7 @@ export function emitLocalArrayRead(
       if (read.element.kind === "bool") {
         const nonzero = B.tmp();
         bits = B.tmp();
-        B.line(`${nonzero} = icmp ne i64 ${value}, 0`);
+        B.line(`${nonzero} = icmp ne i8 ${value}, 0`);
         B.line(`${bits} = zext i1 ${nonzero} to i64`);
       }
       B.line(`store i64 ${bits}, ptr ${payload}`);
@@ -684,7 +688,7 @@ export function emitLocalArrayRead(
       no,
       slow,
       join,
-      scalar ? "i64" : "ptr",
+      !scalar ? "ptr" : read.element.kind === "bool" ? "i8" : "i64",
     );
   else B.br(slow);
   B.startBlock(slow);
diff --git a/packages/compiler/src/backend/runtime-abi.ts b/packages/compiler/src/backend/runtime-abi.ts
index 131f1ae20..2d3550916 100644
--- a/packages/compiler/src/backend/runtime-abi.ts
+++ b/packages/compiler/src/backend/runtime-abi.ts
@@ -2,8 +2,8 @@
  * of the link contract: a program object keeps this symbol undefined and the
  * matching runtime defines it, so incompatible manual links fail before the
  * program can start. */
-export const RUNTIME_ABI_VERSION = 17 as const;
-export const RUNTIME_ABI_MARKER = "scr_runtime_abi_v17" as const;
+export const RUNTIME_ABI_VERSION = 18 as const;
+export const RUNTIME_ABI_MARKER = "scr_runtime_abi_v18" as const;
 
 /** Object consumption is useful today, but the complete scr_* surface may
  * still change before 1.0. The versioned marker prevents accidental mixing;
diff --git a/packages/compiler/src/backend/runtime-pack-targets.test.ts b/packages/compiler/src/backend/runtime-pack-targets.test.ts
index f35905e76..c8c76972a 100644
--- a/packages/compiler/src/backend/runtime-pack-targets.test.ts
+++ b/packages/compiler/src/backend/runtime-pack-targets.test.ts
@@ -52,7 +52,7 @@ async function fixture(target: NativeTargetSpec, workers = false) {
       object_format: target.objectFormat,
       minimum_os: target.minimumOs,
     },
-    runtime_abi: { version: 17, marker: "scr_runtime_abi_v17" },
+    runtime_abi: { version: 18, marker: "scr_runtime_abi_v18" },
     compiler: { command: "fixture", identity: "fixture", target: target.llvmTriple },
     macros: { executable: [], excluded: ["SCR_LIB"], sanitizer: "external-toolchain-required" },
     flavors: {
diff --git a/packages/compiler/src/backend/runtime-pack.test.ts b/packages/compiler/src/backend/runtime-pack.test.ts
index f75d46c48..155f21d59 100644
--- a/packages/compiler/src/backend/runtime-pack.test.ts
+++ b/packages/compiler/src/backend/runtime-pack.test.ts
@@ -116,7 +116,7 @@ async function fixture(target: NativeTargetSpec = MACOS_ARM64_TARGET) {
       object_format: target.objectFormat,
       minimum_os: target.minimumOs,
     },
-    runtime_abi: { version: 17, marker: "scr_runtime_abi_v17" },
+    runtime_abi: { version: 18, marker: "scr_runtime_abi_v18" },
     compiler: {
       command: "clang",
       identity: "fixture clang",
diff --git a/packages/compiler/test/native-codegen-integration.test.ts b/packages/compiler/test/native-codegen-integration.test.ts
index 56fd2c8a4..ca1b86d36 100644
--- a/packages/compiler/test/native-codegen-integration.test.ts
+++ b/packages/compiler/test/native-codegen-integration.test.ts
@@ -312,7 +312,7 @@ ${[0, 1, 2, 3, 4, 5].map((index) => `  call void @add${index}(i64 ${index + 2})`
     expect(await symbols(helperObject, ["-u"])).toEqual(await symbols(clangObject, ["-u"]));
     expect(await symbols(helperObject, ["-gU"])).toEqual(await symbols(clangObject, ["-gU"]));
     expect(await symbols(helperObject, ["-gU"])).toEqual(["0000000000000000 T _main"]);
-    expect(await symbols(helperObject, ["-u"])).toContain("_scr_runtime_abi_v17");
+    expect(await symbols(helperObject, ["-u"])).toContain("_scr_runtime_abi_v18");
 
     const clangExe = join(dir, "clang-program");
     const linkDriver = join(dir, "clang-driver.c");
@@ -646,7 +646,7 @@ ${[0, 1, 2, 3, 4, 5].map((index) => `  call void @add${index}(i64 ${index + 2})`
       .split("\n");
     expect(undefinedSymbols).toContain("_sf_scale");
     expect(undefinedSymbols).toContain("_sf_callback_mix");
-    expect(undefinedSymbols).toContain("_scr_runtime_abi_v17");
+    expect(undefinedSymbols).toContain("_scr_runtime_abi_v18");
   });
 
   test("helper and clang object paths retain the same outbound FFI C ABI", async () => {
diff --git a/packages/compiler/test/ts7/baselines/order-parity/tests/corpus/5080-boolean-array-storage.ts.json b/packages/compiler/test/ts7/baselines/order-parity/tests/corpus/5080-boolean-array-storage.ts.json
new file mode 100644
index 000000000..d1c7860b4
--- /dev/null
+++ b/packages/compiler/test/ts7/baselines/order-parity/tests/corpus/5080-boolean-array-storage.ts.json
@@ -0,0 +1,6 @@
+{
+ "order": [
+  "<repo>/tests/corpus/5080-boolean-array-storage.ts"
+ ],
+ "diags": []
+}
diff --git a/packages/compiler/test/ts7/baselines/order-parity/tests/corpus/5081-boolean-array-dynamic-views.ts.json b/packages/compiler/test/ts7/baselines/order-parity/tests/corpus/5081-boolean-array-dynamic-views.ts.json
new file mode 100644
index 000000000..ff120dad6
--- /dev/null
+++ b/packages/compiler/test/ts7/baselines/order-parity/tests/corpus/5081-boolean-array-dynamic-views.ts.json
@@ -0,0 +1,6 @@
+{
+ "order": [
+  "<repo>/tests/corpus/5081-boolean-array-dynamic-views.ts"
+ ],
+ "diags": []
+}
diff --git a/packages/runtime-darwin-arm64/scripts/build.mjs b/packages/runtime-darwin-arm64/scripts/build.mjs
index 43cc9438a..df08dcde0 100644
--- a/packages/runtime-darwin-arm64/scripts/build.mjs
+++ b/packages/runtime-darwin-arm64/scripts/build.mjs
@@ -306,7 +306,7 @@ async function build() {
       package: packageManifest.name,
       version: packageManifest.version,
       target: RUNTIME_PACK_MATRIX.target,
-      runtime_abi: { version: 17, marker: "scr_runtime_abi_v17" },
+      runtime_abi: { version: 18, marker: "scr_runtime_abi_v18" },
       compiler: {
         command: compiler,
         identity: compilerVersion,
diff --git a/packages/runtime-pack-common/scripts/build.mjs b/packages/runtime-pack-common/scripts/build.mjs
index 5f5364aff..518d7ced6 100644
--- a/packages/runtime-pack-common/scripts/build.mjs
+++ b/packages/runtime-pack-common/scripts/build.mjs
@@ -402,7 +402,7 @@ async function build() {
       package: packageManifest.name,
       version: packageManifest.version,
       target: matrix.target,
-      runtime_abi: { version: 17, marker: "scr_runtime_abi_v17" },
+      runtime_abi: { version: 18, marker: "scr_runtime_abi_v18" },
       compiler: {
         command: compiler,
         identity: compilerVersion,
diff --git a/packages/runtime/src/scr_array.c b/packages/runtime/src/scr_array.c
index 6b26332e9..8cee62491 100644
--- a/packages/runtime/src/scr_array.c
+++ b/packages/runtime/src/scr_array.c
@@ -165,8 +165,19 @@ typedef struct {
   ScrArrSparseSlot *sparse;
   size_t sparse_len;
   size_t sparse_cap;
+  uint8_t elem; /* the owning array's element kind: the dense slot width */
 } ScrArrStorage;
 
+static uint64_t scr_arr_storage_get(const ScrArrStorage *s, size_t index) {
+  if (s->elem == SCR_ELEM_BOOL) return ((const uint8_t *)s->data)[index];
+  return s->data[index];
+}
+
+/* The address of dense slot `index` (<= cap). */
+static uint8_t *scr_arr_dense_at(const ScrArr *a, size_t index) {
+  return (uint8_t *)a->data + index * scr_arr_slot_width(a);
+}
+
 static size_t scr_arr_sparse_lower_bound(const ScrArrSparseSlot *slots,
                                          size_t n, size_t index) {
   size_t lo = 0, hi = n;
@@ -280,8 +291,10 @@ static void scr_arr_sparse_drop_prefix(ScrArr *a, size_t count) {
   x->sparse_len -= count;
 }
 
-/* Bytes per dense slot: the 8-byte value and its state byte. */
+/* Bytes per dense slot: the value and its state byte. Boolean arrays
+ * store a one-byte value. */
 #define SCR_ARR_SLOT_BYTES (sizeof(uint64_t) + sizeof(uint8_t))
+#define SCR_ARR_BOOL_SLOT_BYTES (2 * sizeof(uint8_t))
 /* Appends to an array smaller than this grow straight to it. */
 #define SCR_ARR_MIN_GROWTH ((size_t)4)
 
@@ -297,13 +310,15 @@ typedef enum {
  * literal keeps a 16-byte block rather than reserving 4 slots in a 48-byte
  * one. Growth keeps its own sequence (4, 8, 16, ...): rounding it as well
  * moves array storage into other size classes and measured slower. */
-static size_t scr_arr_block_cap(size_t cap) {
+static size_t scr_arr_block_cap(size_t cap, bool bools) {
 #if SCR_SMALL_ALLOC
-  size_t bytes = cap * SCR_ARR_SLOT_BYTES;
+  size_t bytes = cap * (bools ? SCR_ARR_BOOL_SLOT_BYTES : SCR_ARR_SLOT_BYTES);
   if (bytes <= SCR_SA_MAX) {
     size_t block = (bytes + SCR_SA_STEP - 1) / SCR_SA_STEP * SCR_SA_STEP;
-    return block / SCR_ARR_SLOT_BYTES;
+    return bools ? block / SCR_ARR_BOOL_SLOT_BYTES : block / SCR_ARR_SLOT_BYTES;
   }
+#else
+  (void)bools;
 #endif
   return cap;
 }
@@ -317,9 +332,12 @@ static size_t scr_arr_block_cap(size_t cap) {
 static void scr_arr_grow_dense_to(ScrArr *a, size_t need, ScrArrGrow mode) {
   if (need <= a->cap) return;
   if (need > SCR_ARR_MAX_LENGTH) scr_arr_oom();
+  bool bools = a->elem == SCR_ELEM_BOOL;
+  size_t width = scr_arr_slot_width(a);
+  size_t unit = bools ? SCR_ARR_BOOL_SLOT_BYTES : SCR_ARR_SLOT_BYTES;
   size_t cap;
   if (mode == SCR_ARR_GROW_EXACT) {
-    cap = scr_arr_block_cap(need);
+    cap = scr_arr_block_cap(need, bools);
   } else {
     cap = a->cap < SCR_ARR_MIN_GROWTH ? SCR_ARR_MIN_GROWTH : a->cap;
     while (cap < need) {
@@ -332,13 +350,13 @@ static void scr_arr_grow_dense_to(ScrArr *a, size_t need, ScrArrGrow mode) {
      * appends benefit from geometric headroom. */
     if (mode == SCR_ARR_GROW_SPAN && need > SCR_ARR_DENSE_LIMIT) cap = need;
   }
-  if (cap > SIZE_MAX / SCR_ARR_SLOT_BYTES) scr_arr_oom();
+  if (cap > SIZE_MAX / SCR_ARR_SLOT_BYTES) scr_arr_oom(); /* bounds both slot sizes */
   if (!a->data && SCR_AX(a)->sparse_len == 0) {
     /* First storage: nothing to copy, move or absorb. */
-    uint64_t *data = scr_mem_alloc(cap * SCR_ARR_SLOT_BYTES);
+    uint64_t *data = scr_mem_alloc(cap * unit);
     if (!data) scr_arr_oom();
     a->data = data;
-    a->present = (uint8_t *)(data + cap);
+    a->present = (uint8_t *)data + cap * width;
     memset(a->present, SCR_ARR_HOLE, cap);
     a->cap = cap;
     return;
@@ -346,17 +364,17 @@ static void scr_arr_grow_dense_to(ScrArr *a, size_t need, ScrArrGrow mode) {
   /* Values and states share one allocation. Move the old states before
    * publishing the larger capacity: their previous location is now part
    * of the value area. Existing data and state pointers keep their ABI. */
-  uint64_t *data = scr_mem_realloc(a->data, cap * SCR_ARR_SLOT_BYTES);
+  uint64_t *data = scr_mem_realloc(a->data, cap * unit);
   if (!data) scr_arr_oom();
-  uint8_t *present = (uint8_t *)(data + cap);
-  if (a->cap) memmove(present, (uint8_t *)(data + a->cap), a->cap);
+  uint8_t *present = (uint8_t *)data + cap * width;
+  if (a->cap) memmove(present, (uint8_t *)data + a->cap * width, a->cap);
   memset(present + a->cap, SCR_ARR_HOLE, cap - a->cap);
   a->data = data;
   a->present = present;
   a->cap = cap;
   size_t moved = scr_arr_sparse_lower_bound(SCR_AX(a)->sparse, SCR_AX(a)->sparse_len, cap);
   for (size_t i = 0; i < moved; i++) {
-    a->data[SCR_AX(a)->sparse[i].index] = SCR_AX(a)->sparse[i].slot;
+    scr_arr_dense_put(a, SCR_AX(a)->sparse[i].index, SCR_AX(a)->sparse[i].slot);
     a->present[SCR_AX(a)->sparse[i].index] = SCR_AX(a)->sparse[i].state;
   }
   scr_arr_sparse_drop_prefix(a, moved);
@@ -416,7 +434,7 @@ static uint8_t scr_arr_state_at_storage(const ScrArrStorage *s, size_t index,
                                          uint64_t *out) {
   if (index < s->cap) {
     if (s->present[index] == SCR_ARR_HOLE) return SCR_ARR_HOLE;
-    if (out) *out = s->data[index];
+    if (out) *out = scr_arr_storage_get(s, index);
     return s->present[index];
   }
   size_t pos = scr_arr_sparse_lower_bound(s->sparse, s->sparse_len, index);
@@ -431,14 +449,14 @@ static bool scr_arr_slot_at_storage(const ScrArrStorage *s, size_t index,
 }
 
 static bool scr_arr_slot_at(const ScrArr *a, size_t index, uint64_t *out) {
-  ScrArrStorage s = {
-      a->data, a->present, a->cap, SCR_AX(a)->sparse, SCR_AX(a)->sparse_len, SCR_AX(a)->sparse_cap};
+  ScrArrStorage s = {a->data, a->present, a->cap, SCR_AX(a)->sparse,
+                     SCR_AX(a)->sparse_len, SCR_AX(a)->sparse_cap, a->elem};
   return scr_arr_slot_at_storage(&s, index, out);
 }
 
 static uint8_t scr_arr_state_at(const ScrArr *a, size_t index, uint64_t *out) {
-  ScrArrStorage s = {
-      a->data, a->present, a->cap, SCR_AX(a)->sparse, SCR_AX(a)->sparse_len, SCR_AX(a)->sparse_cap};
+  ScrArrStorage s = {a->data, a->present, a->cap, SCR_AX(a)->sparse,
+                     SCR_AX(a)->sparse_len, SCR_AX(a)->sparse_cap, a->elem};
   return scr_arr_state_at_storage(&s, index, out);
 }
 
@@ -449,7 +467,7 @@ static void scr_arr_store_state_owned(ScrArr *a, size_t index, uint64_t slot,
   if (state != SCR_ARR_VALUE) scr_arr_cut(a, index);
   scr_arr_grow_for_index(a, index);
   if (index < a->cap) {
-    a->data[index] = slot;
+    scr_arr_dense_put(a, index, slot);
     a->present[index] = state;
     return;
   }
@@ -472,9 +490,9 @@ static bool scr_arr_take_state(ScrArr *a, size_t index, uint64_t *out,
     uint8_t state = a->present[index];
     if (state == SCR_ARR_HOLE) return false;
     scr_arr_cut(a, index);
-    *out = a->data[index];
+    *out = scr_arr_dense_get(a, index);
     a->present[index] = SCR_ARR_HOLE;
-    a->data[index] = 0;
+    scr_arr_dense_put(a, index, 0);
     if (state_out) *state_out = state;
     return true;
   }
@@ -498,8 +516,8 @@ static void scr_arr_replace_state_owned(ScrArr *a, size_t index, uint64_t slot,
   if (state != SCR_ARR_VALUE) scr_arr_cut(a, index);
   if (index < a->cap) {
     old_state = a->present[index];
-    if (old_state == SCR_ARR_VALUE) old = a->data[index];
-    a->data[index] = slot;
+    if (old_state == SCR_ARR_VALUE) old = scr_arr_dense_get(a, index);
+    scr_arr_dense_put(a, index, slot);
     a->present[index] = state;
   } else {
     size_t pos = scr_arr_sparse_lower_bound(SCR_AX(a)->sparse, SCR_AX(a)->sparse_len, index);
@@ -610,9 +628,8 @@ static bool scr_arr_prop_delete(ScrArr *a, double key) {
 }
 
 static ScrArrStorage scr_arr_take_storage(ScrArr *a) {
-  ScrArrStorage old = {
-      a->data, a->present, a->cap, SCR_AX(a)->sparse, SCR_AX(a)->sparse_len,
-      SCR_AX(a)->sparse_cap};
+  ScrArrStorage old = {a->data, a->present, a->cap, SCR_AX(a)->sparse,
+                       SCR_AX(a)->sparse_len, SCR_AX(a)->sparse_cap, a->elem};
   a->data = NULL;
   a->present = NULL;
   a->cap = 0;
@@ -641,7 +658,7 @@ static bool scr_arr_is_dense(const ScrArr *a) {
 static void scr_arr_move_dense(ScrArr *a, size_t to, size_t from, size_t count) {
   if (count == 0) return;
   scr_arr_cut(a, to);
-  memmove(a->data + to, a->data + from, count * sizeof(*a->data));
+  memmove(scr_arr_dense_at(a, to), scr_arr_dense_at(a, from), count * scr_arr_slot_width(a));
   memmove(a->present + to, a->present + from, count);
 }
 
@@ -649,24 +666,26 @@ static void scr_arr_clear_dense(ScrArr *a, size_t from, size_t count) {
   if (count == 0) return;
   scr_arr_cut(a, from);
   memset(a->present + from, SCR_ARR_HOLE, count);
-  memset(a->data + from, 0, count * sizeof(*a->data));
+  memset(scr_arr_dense_at(a, from), 0, count * scr_arr_slot_width(a));
 }
 
 /* The destination owns no values in this range. Only retain adapters run
  * while copying references: no release can collect an intermediate copy.
- * Self-appends use disjoint ranges after reserving capacity. */
+ * Self-appends use disjoint ranges after reserving capacity. Both arrays
+ * hold the same element kind, so their slots have the same width. */
 static void scr_arr_copy_dense(ScrArr *dst, size_t to, const ScrArr *src,
                                 size_t from, size_t count, bool reverse,
                                 bool materialize_holes) {
   if (count == 0) return;
   scr_arr_cut(dst, to);
   if (!reverse) {
-    memcpy(dst->data + to, src->data + from, count * sizeof(*dst->data));
+    memcpy(scr_arr_dense_at(dst, to), scr_arr_dense_at(src, from),
+           count * scr_arr_slot_width(dst));
     memcpy(dst->present + to, src->present + from, count);
   } else {
     for (size_t i = 0; i < count; i++) {
       size_t source = from + count - i - 1;
-      memcpy(dst->data + to + i, src->data + source, sizeof(*dst->data));
+      scr_arr_dense_put(dst, to + i, scr_arr_dense_get(src, source));
       dst->present[to + i] = src->present[source];
     }
   }
@@ -676,12 +695,21 @@ static void scr_arr_copy_dense(ScrArr *dst, size_t to, const ScrArr *src,
     if (dst->present[i] == SCR_ARR_VALUE) {
       if (refs) dst->data[i] = scr_elem_retain_slot(src, dst->data[i]);
     } else if (materialize_holes && dst->present[i] == SCR_ARR_HOLE) {
-      dst->data[i] = 0;
+      scr_arr_dense_put(dst, i, 0);
       dst->present[i] = SCR_ARR_UNDEFINED;
     }
   }
 }
 
+/* Store `count` 8-byte value slots into dense storage starting at `to`. */
+static void scr_arr_put_slots(ScrArr *a, size_t to, size_t count, const uint64_t *slots) {
+  if (a->elem != SCR_ELEM_BOOL) {
+    memcpy(a->data + to, slots, count * sizeof(*slots));
+    return;
+  }
+  for (size_t i = 0; i < count; i++) scr_arr_dense_put(a, to + i, slots[i]);
+}
+
 ScrArr *scr_arr_new(ScrElemKind elem, size_t initial_cap) {
   ScrArr *a = scr_mem_alloc(sizeof(ScrArr));
   if (!a) scr_arr_oom();
@@ -926,9 +954,10 @@ void scr_arr_set_len(ScrArr *a, double length) {
         }
       }
     } else {
-      for (size_t i = dense_stop; i < dense_end; i++) {
-        a->present[i] = 0;
-        a->data[i] = 0;
+      if (dense_stop < dense_end) {
+        memset(a->present + dense_stop, 0, dense_end - dense_stop);
+        memset(scr_arr_dense_at(a, dense_stop), 0,
+               (dense_end - dense_stop) * scr_arr_slot_width(a));
       }
       if (a->ext) a->ext->sparse_len = scr_arr_sparse_lower_bound(a->ext->sparse, a->ext->sparse_len, next);
     }
@@ -1053,7 +1082,8 @@ static ScrArr *scr_arr_fill_slot(ScrArr *a, uint64_t slot, uint8_t state,
   if (scr_arr_dense_ok(a, until, until - from)) scr_arr_grow_dense(a, until);
   if (!refs && until <= a->cap) {
     if (state != SCR_ARR_VALUE) scr_arr_cut(a, from);
-    for (size_t i = from; i < until; i++) a->data[i] = slot;
+    if (a->elem == SCR_ELEM_BOOL) memset(scr_arr_dense_at(a, from), (int)slot, until - from);
+    else for (size_t i = from; i < until; i++) a->data[i] = slot;
     memset(a->present + from, state, until - from);
   } else {
     for (size_t i = from; i < until; i++) {
@@ -1241,8 +1271,8 @@ bool scr_arr_index_eq(const ScrArr *a, double i, const ScrArr *b, double j) {
     bool ap = ai < a->len && a->present[ai] == SCR_ARR_VALUE;
     bool bp = bi < b->len && b->present[bi] == SCR_ARR_VALUE;
     if (!ap || !bp) return ap == bp;
-    av = a->data[ai];
-    bv = b->data[bi];
+    av = scr_arr_dense_get(a, ai);
+    bv = scr_arr_dense_get(b, bi);
   } else {
     bool ap = scr_arr_read_value(a, i, &av);
     bool bp = scr_arr_read_value(b, j, &bv);
@@ -1352,7 +1382,7 @@ double scr_arr_push_many(ScrArr *a, size_t count, const uint64_t *slots) {
   if (scr_arr_dense_ok(a, next, count)) {
     scr_arr_grow_dense(a, next);
     if (count) {
-      memcpy(a->data + from, slots, count * sizeof(*slots));
+      scr_arr_put_slots(a, from, count, slots);
       memset(a->present + from, SCR_ARR_VALUE, count);
     }
   } else {
@@ -1467,7 +1497,7 @@ static double scr_arr_unshift_slot(ScrArr *a, uint64_t slot) {
     /* The prefix moves up behind the new first value. */
     size_t f = a->filled;
     scr_arr_move_dense(a, 1, 0, old_len);
-    a->data[0] = slot;
+    scr_arr_dense_put(a, 0, slot);
     a->present[0] = SCR_ARR_VALUE;
     a->len++;
     a->filled = (uint32_t)(f + 1);
@@ -1477,7 +1507,8 @@ static double scr_arr_unshift_slot(ScrArr *a, uint64_t slot) {
   a->len = 0;
   scr_arr_store_owned(a, 0, slot);
   for (size_t i = 0; i < old.cap; i++) {
-    if (old.present[i]) scr_arr_store_state_owned(a, i + 1, old.data[i], old.present[i]);
+    if (old.present[i])
+      scr_arr_store_state_owned(a, i + 1, scr_arr_storage_get(&old, i), old.present[i]);
   }
   for (size_t i = 0; i < old.sparse_len; i++) {
     scr_arr_store_state_owned(a, old.sparse[i].index + 1, old.sparse[i].slot,
@@ -1509,7 +1540,7 @@ double scr_arr_unshift_many(ScrArr *a, size_t count, const uint64_t *slots) {
     scr_arr_grow_dense(a, old_len + count);
     size_t f = a->filled;
     scr_arr_move_dense(a, count, 0, old_len);
-    memcpy(a->data, slots, count * sizeof(*slots));
+    scr_arr_put_slots(a, 0, count, slots);
     memset(a->present, SCR_ARR_VALUE, count);
     a->len = old_len + count;
     a->filled = (uint32_t)(f + count);
@@ -1519,7 +1550,7 @@ double scr_arr_unshift_many(ScrArr *a, size_t count, const uint64_t *slots) {
     for (size_t i = 0; i < count; i++) scr_arr_store_owned(a, i, slots[i]);
     for (size_t i = 0; i < old.cap; i++) {
       if (old.present[i] != SCR_ARR_HOLE)
-        scr_arr_store_state_owned(a, i + count, old.data[i], old.present[i]);
+        scr_arr_store_state_owned(a, i + count, scr_arr_storage_get(&old, i), old.present[i]);
     }
     for (size_t i = 0; i < old.sparse_len; i++) {
       scr_arr_store_state_owned(a, old.sparse[i].index + count,
@@ -1562,7 +1593,7 @@ double scr_arr_unshift_spread(ScrArr *a, const ScrArr *src) {
   }
   for (size_t i = 0; i < old.cap; i++) {
     if (old.present[i] != SCR_ARR_HOLE) {
-      scr_arr_store_state_owned(a, i + add, old.data[i], old.present[i]);
+      scr_arr_store_state_owned(a, i + add, scr_arr_storage_get(&old, i), old.present[i]);
     }
   }
   for (size_t i = 0; i < old.sparse_len; i++) {
@@ -1582,10 +1613,9 @@ ScrArr *scr_arr_reverse(ScrArr *a) {
     if (a->filled != len) scr_arr_cut(a, 0);
     for (size_t i = 0; i < len / 2; i++) {
       size_t j = len - i - 1;
-      uint64_t slot;
-      memcpy(&slot, a->data + i, sizeof slot);
-      memcpy(a->data + i, a->data + j, sizeof slot);
-      memcpy(a->data + j, &slot, sizeof slot);
+      uint64_t slot = scr_arr_dense_get(a, i);
+      scr_arr_dense_put(a, i, scr_arr_dense_get(a, j));
+      scr_arr_dense_put(a, j, slot);
       uint8_t state = a->present[i];
       a->present[i] = a->present[j];
       a->present[j] = state;
@@ -1596,7 +1626,8 @@ ScrArr *scr_arr_reverse(ScrArr *a) {
   ScrArrStorage old = scr_arr_take_storage(a);
   a->len = len;
   for (size_t i = 0; i < old.cap; i++) {
-    if (old.present[i]) scr_arr_store_state_owned(a, len - 1 - i, old.data[i], old.present[i]);
+    if (old.present[i])
+      scr_arr_store_state_owned(a, len - 1 - i, scr_arr_storage_get(&old, i), old.present[i]);
   }
   for (size_t i = 0; i < old.sparse_len; i++) {
     scr_arr_store_state_owned(a, len - 1 - old.sparse[i].index,
@@ -1645,7 +1676,7 @@ uint8_t scr_arr_shift_state(ScrArr *a, uint64_t *slot_out) {
   size_t old_len = a->len;
   if (scr_arr_is_dense(a)) {
     uint8_t state = a->present[0];
-    if (slot_out && state == SCR_ARR_VALUE) *slot_out = a->data[0];
+    if (slot_out && state == SCR_ARR_VALUE) *slot_out = scr_arr_dense_get(a, 0);
     /* The prefix slides down with the tail. */
     size_t f = a->filled;
     scr_arr_move_dense(a, 0, 1, old_len - 1);
@@ -1659,7 +1690,8 @@ uint8_t scr_arr_shift_state(ScrArr *a, uint64_t *slot_out) {
   uint8_t state = scr_arr_state_at_storage(&old, 0, &s);
   a->len = 0;
   for (size_t i = 1; i < old.cap; i++) {
-    if (old.present[i]) scr_arr_store_state_owned(a, i - 1, old.data[i], old.present[i]);
+    if (old.present[i])
+      scr_arr_store_state_owned(a, i - 1, scr_arr_storage_get(&old, i), old.present[i]);
   }
   for (size_t i = 0; i < old.sparse_len; i++) {
     if (old.sparse[i].index > 0) {
@@ -1716,7 +1748,7 @@ ScrArr *scr_arr_splice(ScrArr *a, double start, double deleteCount) {
   size_t old_len = a->len;
   if (scr_arr_is_dense(a)) {
     if (n) {
-      memcpy(out->data, a->data + from, n * sizeof(*a->data));
+      memcpy(out->data, scr_arr_dense_at(a, from), n * scr_arr_slot_width(a));
       memcpy(out->present, a->present + from, n);
     }
     scr_arr_refill(out);
@@ -1735,16 +1767,17 @@ ScrArr *scr_arr_splice(ScrArr *a, double start, double deleteCount) {
   for (size_t i = 0; i < old.cap; i++) {
     if (old.present[i] == SCR_ARR_HOLE) continue;
     uint8_t state = old.present[i];
+    uint64_t slot = scr_arr_storage_get(&old, i);
     size_t dst;
     if (i < from) {
       dst = i;
-      scr_arr_store_state_owned(a, dst, old.data[i], state);
+      scr_arr_store_state_owned(a, dst, slot, state);
     } else if (i >= from + n) {
       dst = i - n;
-      scr_arr_store_state_owned(a, dst, old.data[i], state);
+      scr_arr_store_state_owned(a, dst, slot, state);
     } else {
       dst = i - from;
-      scr_arr_store_state_owned(out, dst, old.data[i], state);
+      scr_arr_store_state_owned(out, dst, slot, state);
     }
   }
   for (size_t j = 0; j < old.sparse_len; j++) {
@@ -1790,7 +1823,7 @@ ScrArr *scr_arr_splice_insert(ScrArr *a, double start, double deleteCount,
         ? scr_arr_new_ops(a->ops, n)
         : scr_arr_new(a->elem, n);
     if (n) {
-      memcpy(removed->data, a->data + from, n * sizeof(*a->data));
+      memcpy(removed->data, scr_arr_dense_at(a, from), n * scr_arr_slot_width(a));
       memcpy(removed->present, a->present + from, n);
     }
     removed->len = n;
@@ -1858,7 +1891,7 @@ void scr_arr_splice_drop(ScrArr *a, double start, double deleteCount, size_t cou
   bool packed = a->filled == old_len;
   scr_arr_move_dense(a, from + count, from + n, old_len - from - n);
   if (count) {
-    memcpy(a->data + from, slots, count * sizeof(*slots));
+    scr_arr_put_slots(a, from, count, slots);
     memset(a->present + from, SCR_ARR_VALUE, count);
   }
   if (next_len < old_len) scr_arr_clear_dense(a, next_len, old_len - next_len);
@@ -1927,7 +1960,7 @@ double scr_arr_index_of_f64(ScrArr *a, double v) {
 
 double scr_arr_index_of_bool(ScrArr *a, bool v) {
   for (size_t i = 0; i < a->cap && i < a->len; i++) {
-    if (a->present[i] == SCR_ARR_VALUE && ((a->data[i] != 0) == v)) return (double)i;
+    if (a->present[i] == SCR_ARR_VALUE && ((scr_arr_dense_get(a, i) != 0) == v)) return (double)i;
   }
   for (size_t i = 0; i < SCR_AX(a)->sparse_len; i++) {
     if (SCR_AX(a)->sparse[i].state == SCR_ARR_VALUE && (SCR_AX(a)->sparse[i].slot != 0) == v) return (double)SCR_AX(a)->sparse[i].index;
@@ -2042,7 +2075,7 @@ ScrArr *scr_arr_slice(ScrArr *a, double start, double end) {
     if (i < from || a->present[i] == SCR_ARR_HOLE) continue;
     uint8_t state = a->present[i];
     scr_arr_store_state_owned(out, i - from,
-                              scr_arr_retain_state_slot(a, a->data[i], state),
+                              scr_arr_retain_state_slot(a, scr_arr_dense_get(a, i), state),
                               state);
   }
   for (size_t i = 0; i < SCR_AX(a)->sparse_len; i++) {
@@ -2179,7 +2212,7 @@ ScrArr *scr_arr_sort_primitive(ScrArr *a, bool copy) {
   if (count && !values) scr_arr_oom();
   size_t used = 0;
   for (size_t i = 0; i < dense; i++)
-    if (a->present[i] == SCR_ARR_VALUE) values[used++] = a->data[i];
+    if (a->present[i] == SCR_ARR_VALUE) values[used++] = scr_arr_dense_get(a, i);
   for (size_t i = 0; i < SCR_AX(a)->sparse_len; i++)
     if (SCR_AX(a)->sparse[i].state == SCR_ARR_VALUE) values[used++] = SCR_AX(a)->sparse[i].slot;
   ScrNumberSortKey *numbers = NULL;
@@ -2203,12 +2236,13 @@ ScrArr *scr_arr_sort_primitive(ScrArr *a, bool copy) {
   ScrArr *out = copy ? scr_arr_new(a->elem, present) : a;
   if (copy) out->len = a->len;
   if (!copy && scr_arr_is_dense(a)) {
-    if (count) memcpy(a->data, values, count * sizeof(*values));
+    if (count) scr_arr_put_slots(a, 0, count, values);
     scr_arr_cut(a, count);
     memset(a->present, SCR_ARR_VALUE, count);
     memset(a->present + count, SCR_ARR_UNDEFINED, undefined_count);
     scr_arr_clear_dense(a, count + undefined_count, a->len - count - undefined_count);
-    if (undefined_count) memset(a->data + count, 0, undefined_count * sizeof(*a->data));
+    if (undefined_count)
+      memset(scr_arr_dense_at(a, count), 0, undefined_count * scr_arr_slot_width(a));
   } else {
     ScrArrStorage old = {0};
     if (!copy) old = scr_arr_take_storage(a);
@@ -2457,7 +2491,7 @@ void scr_arr_sort_values(ScrArr *a, double count, ScrClosure *f, uint32_t arity)
   /* A permutation of the same owned slots: no reference count changes. */
   for (size_t i = 0; sorted && i < n; i++) {
     if (i < a->cap) {
-      a->data[i] = src[i];
+      scr_arr_dense_put(a, i, src[i]);
     } else {
       size_t pos = scr_arr_sparse_lower_bound(SCR_AX(a)->sparse, SCR_AX(a)->sparse_len, i);
       SCR_AX(a)->sparse[pos].slot = src[i];
diff --git a/packages/runtime/src/scr_console.c b/packages/runtime/src/scr_console.c
index 035c42318..ca834f754 100644
--- a/packages/runtime/src/scr_console.c
+++ b/packages/runtime/src/scr_console.c
@@ -126,7 +126,7 @@ static void scr_enable_utf8_console_output(void) {
 }
 #endif
 
-void scr_runtime_abi_v17(void) {}
+void scr_runtime_abi_v18(void) {}
 
 void scr_init(void) {
 #ifdef SCR_WORKERS
diff --git a/packages/runtime/src/scr_lib.c b/packages/runtime/src/scr_lib.c
index 6b2222b74..4c3916b3c 100644
--- a/packages/runtime/src/scr_lib.c
+++ b/packages/runtime/src/scr_lib.c
@@ -7625,7 +7625,7 @@ ScrArr *scr_set_to_arr_bool(const ScrMap *s) {
   if (scr_set_to_arr_sparse(s, out)) return out;
   for (size_t i = 0; i < s->nentries; i++) {
     if (!s->entries[i].hash) continue;
-    out->data[out->len] = s->entries[i].key;
+    scr_arr_dense_put(out, out->len, s->entries[i].key != 0);
     out->present[out->len++] = SCR_ARR_VALUE;
   }
   return out;
diff --git a/packages/runtime/src/scr_map.c b/packages/runtime/src/scr_map.c
index 2f8582c56..5b6e333e8 100644
--- a/packages/runtime/src/scr_map.c
+++ b/packages/runtime/src/scr_map.c
@@ -1462,7 +1462,7 @@ void scr_set_add_all(ScrMap *set, ScrArr *values) {
      * Avoid taking a temporary owner only to release it after insertion.
      * Sparse/missing positions retain the existing accessor semantics. */
     if (i < values->cap && values->present[i] == SCR_ARR_VALUE) {
-      uint64_t key = values->data[i];
+      uint64_t key = scr_arr_dense_get(values, i);
       if (set->key_kind == SCR_MAP_KEY_F64) key = scr_map_f64_bits(scr_map_slot_to_f64(key));
       scr_map_set(set, scr_map_hash_key(set, key), key, 0);
     } else if (values->elem == SCR_ELEM_STR) {
diff --git a/packages/runtime/src/scr_runtime.h b/packages/runtime/src/scr_runtime.h
index 4456f8e8f..b01504305 100644
--- a/packages/runtime/src/scr_runtime.h
+++ b/packages/runtime/src/scr_runtime.h
@@ -74,7 +74,7 @@ void scr_init(void);
 /* Program objects emitted by the bundled LLVM helper reference this symbol.
  * Its versioned spelling makes a mismatched manual runtime link fail before
  * the program can start. */
-void scr_runtime_abi_v17(void);
+void scr_runtime_abi_v18(void);
 void scr_runtime_workers_v8(void);
 
 /* ── the trap funnel (scr_console.c; scr_library.c under -DSCR_LIB) ──────
@@ -1536,8 +1536,9 @@ ScrStr *scr_b64_missing_arg(void);
 
 /* ── arrays ─────────────────────────────────────────────────────────
  * Monomorphic, growable, refcounted. Elements live in 8-byte slots
- * (doubles/bools/pointers via memcpy and casts); `elem` records what the
- * slots hold so the array can release its own reference elements — ScrStr's
+ * (doubles/pointers via memcpy and casts), except SCR_ELEM_BOOL, whose
+ * dense storage holds one byte (0 or 1) per element; `elem` records what
+ * the slots hold so the array can release its own reference elements — ScrStr's
  * layout is untouched (interned-literal statics depend on it), so the
  * element kind lives here instead of in a shared object header. A separate
  * presence bitset distinguishes a hole from a present `undefined`-carrying
@@ -1614,6 +1615,9 @@ typedef struct ScrArr {
   uint32_t filled;
   /* SCR_ELEM_REF only; NULL for every other element kind. */
   const ScrArrOps *ops;
+  /* Dense values: `cap` 8-byte slots, or `cap` bytes for SCR_ELEM_BOOL
+   * (scr_arr_slot_width). Sparse entries and properties keep 8-byte slots
+   * for every kind; a bool slot is 0 or 1 wherever it lives. */
   uint64_t *data;
   uint8_t *present; /* one state per dense slot; heap arrays share data allocation */
   /* Rarely used storage (sparse indices, noncanonical numeric properties,
@@ -1642,6 +1646,22 @@ _Static_assert(offsetof(ScrArr, filled) == 3 * sizeof(size_t) + 4,
 _Static_assert(offsetof(ScrArr, data) == 3 * sizeof(size_t) + 8 + sizeof(void *),
                "inline element reads use ScrArr.data as %ScrArr field 8");
 
+/* Bytes per dense value slot. */
+static inline size_t scr_arr_slot_width(const ScrArr *a) {
+  return a->elem == SCR_ELEM_BOOL ? 1 : sizeof(uint64_t);
+}
+
+/* The dense value at `index` (< cap) as an 8-byte slot, and its store. */
+static inline uint64_t scr_arr_dense_get(const ScrArr *a, size_t index) {
+  if (a->elem == SCR_ELEM_BOOL) return ((const uint8_t *)a->data)[index];
+  return a->data[index];
+}
+
+static inline void scr_arr_dense_put(ScrArr *a, size_t index, uint64_t slot) {
+  if (a->elem == SCR_ELEM_BOOL) ((uint8_t *)a->data)[index] = (uint8_t)slot;
+  else a->data[index] = slot;
+}
+
 /* Lower the hole-free prefix to at most `index`: call before storing a
  * non-value state at `index`, or before moving states into [index, ...). */
 static inline void scr_arr_cut(ScrArr *a, size_t index) {
diff --git a/packages/runtime/test/test_array.c b/packages/runtime/test/test_array.c
index 14c50b5d8..0a05031b5 100644
--- a/packages/runtime/test/test_array.c
+++ b/packages/runtime/test/test_array.c
@@ -1365,6 +1365,107 @@ static void test_filled_prefix(void) {
   scr_arr_release(r);
 }
 
+/* Boolean arrays store one byte per dense element, followed by the states in
+ * the same allocation. Growth moves the states past the larger value area,
+ * densifying absorbs sparse entries into bytes, and every moving and copying
+ * operation keeps values, holes and present undefined in place. */
+static bool bool_layout_ok(const ScrArr *a) {
+  return a->cap == 0 || a->present == (const uint8_t *)a->data + a->cap;
+}
+
+static bool bool_states(const ScrArr *a, const char *want) {
+  if (scr_arr_len(a) != (double)strlen(want)) return false;
+  for (size_t i = 0; want[i]; i++) {
+    double state = scr_arr_state(a, (double)i);
+    char got = state == SCR_ARR_HOLE        ? '_'
+               : state == SCR_ARR_UNDEFINED ? 'u'
+               : scr_arr_get_bool((ScrArr *)a, (double)i) ? 't'
+                                                          : 'f';
+    if (got != want[i]) return false;
+  }
+  return true;
+}
+
+static void test_bool_bytes(void) {
+  ScrArr *a = scr_arr_new(SCR_ELEM_BOOL, 0);
+  for (int i = 0; i < 1000; i++) scr_arr_push_bool(a, i % 3 == 0);
+  check(bool_layout_ok(a) && a->filled == 1000, "bool pushes grow one-byte storage");
+  bool values_ok = true;
+  for (int i = 0; i < 1000; i++) {
+    values_ok &= ((const uint8_t *)a->data)[i] == (i % 3 == 0);
+    values_ok &= scr_arr_get_bool(a, i) == (i % 3 == 0);
+  }
+  check(values_ok, "bool values are stored as 0/1 bytes");
+  check(scr_arr_index_of_bool(a, false) == 1 && scr_arr_includes_bool(a, true),
+        "bool search reads bytes");
+  scr_arr_set_len(a, 5);
+  scr_arr_set_len(a, 8);
+  scr_arr_set_undefined(a, 6);
+  check(bool_layout_ok(a) && bool_states(a, "tfftf_u_"), "bool truncation clears bytes");
+
+  ScrArr *s = scr_arr_new(SCR_ELEM_BOOL, 0);
+  scr_arr_set_bool(s, 0, true);
+  scr_arr_set_bool(s, 3000000, true);
+  scr_arr_set_bool(s, 2000000, false);
+  check(arr_sparse_len(s) == 2 && !scr_arr_get_bool(s, 2000000) && scr_arr_get_bool(s, 3000000),
+        "bool sparse entries keep their values");
+  scr_arr_set_len(s, 2000001);
+  scr_arr_release(scr_arr_fill_bool(s, true, 1, 2000000));
+  check(arr_sparse_len(s) == 0 && bool_layout_ok(s) && !scr_arr_get_bool(s, 2000000) &&
+            scr_arr_get_bool(s, 1999999) && scr_arr_index_of_bool(s, false) == 2000000,
+        "densifying absorbs sparse bool entries into bytes");
+  scr_arr_release(s);
+
+  ScrArr *m = scr_arr_new(SCR_ELEM_BOOL, 0);
+  uint64_t slots[3] = {1, 0, 1};
+  scr_arr_push_many(m, 3, slots);
+  scr_arr_unshift_many(m, 2, slots + 1);
+  check(bool_states(m, "fttft"), "slot buffers narrow to bytes");
+  scr_arr_unshift_bool(m, false);
+  scr_arr_set_undefined(m, 2);
+  scr_arr_delete(m, 4);
+  check(bool_states(m, "ffut_t"), "unshift moves bytes and states together");
+  ScrArr *removed = scr_arr_splice(m, 1, 2);
+  check(bool_states(removed, "fu") && bool_states(m, "ft_t"), "splice moves bytes out");
+  scr_arr_release(removed);
+  ScrArr *items = scr_arr_new(SCR_ELEM_BOOL, 0);
+  scr_arr_push_bool(items, true);
+  scr_arr_push_bool(items, false);
+  scr_arr_release(scr_arr_splice_insert(m, 1, 1, items));
+  check(bool_states(m, "ftf_t"), "splice insertion copies bytes in");
+  scr_arr_splice_drop(m, 0, 1, 2, slots);
+  check(bool_states(m, "tftf_t"), "statement splice stores bytes");
+  scr_arr_release(scr_arr_reverse(m));
+  check(bool_states(m, "t_ftft"), "reverse swaps bytes");
+  ScrArr *copy = scr_arr_slice(m, 1, INFINITY);
+  check(bool_states(copy, "_ftft") && bool_layout_ok(copy), "slice copies bytes");
+  scr_arr_release(copy);
+  copy = scr_arr_to_reversed(m);
+  check(bool_states(copy, "tftfut"), "toReversed copies bytes and materializes holes");
+  scr_arr_release(copy);
+  scr_arr_concat_copy(m, items);
+  scr_arr_push_spread(m, m);
+  check(bool_states(m, "t_ftfttftuftfttf"), "concat keeps holes; self spread reads them as undefined");
+  scr_arr_release(scr_arr_copy_within(m, 0, 2, 5));
+  check(scr_arr_index_eq(m, 0, items, 1) && scr_arr_index_eq(m, 1, items, 0),
+        "copyWithin moves bytes and index equality reads them");
+  check(scr_arr_shift_bool(m) == false && scr_arr_pop_bool(m) == false, "shift and pop read bytes");
+  scr_arr_release(scr_arr_sort_primitive(m, false));
+  bool sorted = true;
+  for (double i = 1; i < scr_arr_len(m); i++)
+    if (scr_arr_state(m, i) == SCR_ARR_VALUE && scr_arr_state(m, i - 1) == SCR_ARR_VALUE)
+      sorted &= scr_arr_get_bool(m, i - 1) <= scr_arr_get_bool(m, i);
+  check(sorted, "default sort orders bytes");
+  ScrStr *sep = scr_str_new(",", 1);
+  ScrStr *joined = scr_arr_join(items, sep);
+  check(strcmp(joined->data, "true,false") == 0, "join reads bytes");
+  scr_str_release(joined);
+  scr_str_release(sep);
+  scr_arr_release(items);
+  scr_arr_release(m);
+  scr_arr_release(a);
+}
+
 int main(int argc, char **argv) {
   if (argc > 1) {
     ScrArr *a = scr_arr_new(SCR_ELEM_F64, 0);
@@ -1410,6 +1511,7 @@ int main(int argc, char **argv) {
   test_bulk_reference_ownership();
   test_splice_drop();
   test_filled_prefix();
+  test_bool_bytes();
 
   fprintf(stderr, "%ld/%ld cases passed\n", total - failed, total);
   return failed == 0 ? 0 : 1;
diff --git a/tests/corpus/5080-boolean-array-storage.ts b/tests/corpus/5080-boolean-array-storage.ts
new file mode 100644
index 000000000..99727ee78
--- /dev/null
+++ b/tests/corpus/5080-boolean-array-storage.ts
@@ -0,0 +1,183 @@
+// boolean[] keeps one byte per element in dense storage. Every array
+// operation reads and writes those bytes: element access below and past the
+// hole-free prefix, growth and truncation, the moving operations (push, pop,
+// shift, unshift, splice, reverse, copyWithin), the copying ones (slice,
+// concat, spread, toReversed, toSpliced, with), fill, sort, search, join,
+// iteration and JSON. Holes and present undefined stay distinct throughout.
+function show(label: string, xs: boolean[]): void {
+  const parts: string[] = [];
+  for (let i = 0; i < xs.length; i++) parts.push(i in xs ? String(xs[i]) : "_");
+  let visits = 0;
+  let trues = 0;
+  xs.forEach((x) => {
+    visits++;
+    if (x) trues++;
+  });
+  console.log(label, xs.length, parts.join(","), "visits", visits, "trues", trues);
+}
+
+function showMaybe(label: string, xs: (boolean | undefined)[]): void {
+  const parts: string[] = [];
+  for (let i = 0; i < xs.length; i++) parts.push(i in xs ? String(xs[i]) : "_");
+  let visits = 0;
+  xs.forEach(() => visits++);
+  console.log(label, xs.length, parts.join(","), "visits", visits);
+}
+
+// A hole at `at` without `delete`: copying a hole deletes the target.
+function punch(xs: boolean[], at: number): void {
+  const n = xs.length;
+  xs.length = n + 1;
+  xs.copyWithin(at, n, n + 1);
+  xs.length = n;
+}
+
+// Element writes inside, at and past the length.
+const a: boolean[] = new Array<boolean>(6).fill(false);
+show("fill", a);
+a[1] = true;
+a[4] = true;
+show("write", a);
+a[8] = true;
+show("write-past", a);
+console.log("in", 6 in a, 7 in a, 8 in a, a[7]);
+a[6] = false;
+a[7] = true;
+show("close-gap", a);
+a.length = 3;
+show("truncate", a);
+a.length = 5;
+show("regrow", a);
+a[3] = true;
+a[4] = false;
+show("rewrite", a);
+
+// Moving operations.
+const b: boolean[] = [true, false, true, true, false, false, true];
+punch(b, 2);
+show("punch", b);
+console.log("push", b.push(true), b.push(false, true));
+show("pushed", b);
+console.log("pop", b.pop(), b.pop(), b.length);
+console.log("shift", b.shift(), b.shift(), b.length);
+show("shifted", b);
+console.log("unshift", b.unshift(false), b.unshift(true, true));
+show("unshifted", b);
+console.log("splice", b.splice(1, 2).join(","));
+show("spliced", b);
+console.log("splice-insert", b.splice(2, 1, true, false, false).join(","));
+show("inserted", b);
+b.splice(0, 2, false);
+show("splice-statement", b);
+b.reverse();
+show("reverse", b);
+b.copyWithin(0, 3, 6);
+show("copyWithin", b);
+const holey: boolean[] = [true, false, true, false];
+punch(holey, 1);
+holey.reverse();
+show("reverse-holes", holey);
+console.log("shift-hole", holey.shift(), holey.shift(), holey.length);
+show("after-shift-hole", holey);
+
+// Copying operations preserve holes, except where iteration reads them.
+const c: boolean[] = [false, true, false, true, true];
+punch(c, 3);
+show("source", c);
+show("slice", c.slice(1, 5));
+show("slice-neg", c.slice(-3));
+show("concat", c.concat([true], c));
+show("spread", [...c, false]);
+show("toReversed", c.toReversed());
+show("toSpliced", c.toSpliced(1, 2, true));
+show("with", c.with(0, true));
+show("from", Array.from(c));
+const d: boolean[] = [];
+d.push(...c);
+show("push-spread", d);
+d.unshift(...c);
+show("unshift-spread", d);
+
+// fill, sort and search.
+const e: boolean[] = new Array<boolean>(10).fill(true, 0, 10);
+e.fill(false, 2, 5);
+e.fill(false, -2);
+show("fill-range", e);
+show("toSorted", e.toSorted());
+e.sort();
+show("sort", e);
+const f: boolean[] = [true, false, false, true];
+f.sort((x, y) => Number(y) - Number(x));
+show("sort-cmp", f);
+console.log("indexOf", e.indexOf(true), e.indexOf(false), e.lastIndexOf(false), [true].indexOf(false));
+console.log("includes", e.includes(true), [false, false].includes(true));
+console.log("find", e.find((x) => x), e.findIndex((x) => x), e.findLast((x) => !x), e.findLastIndex((x) => !x));
+console.log("some/every", e.some((x) => x), e.every((x) => x), [true, true].every((x) => x));
+console.log("count", e.filter((x) => x).length, e.map((x) => (x ? 1 : 0)).join(""));
+console.log("reduce", e.reduce((n, x) => (x ? n + 1 : n), 0));
+
+// Strings, iteration and JSON.
+const g: boolean[] = [true, false, true];
+punch(g, 1);
+console.log("join", g.join(), g.join(" | "), String(g), `${g}`);
+const seen: string[] = [];
+for (const x of g) seen.push(String(x));
+for (const [i, x] of g.entries()) seen.push(`${i}:${x}`);
+for (const i of g.keys()) seen.push(`k${i}`);
+console.log("iterate", seen.join(" "));
+console.log("json", JSON.stringify(g), JSON.stringify([true, false]), JSON.stringify({ flags: [false, true] }));
+const parsed: boolean[] = JSON.parse("[true,false,false,true]") as boolean[];
+parsed.push(true);
+console.log("parsed", parsed.length, parsed[3], parsed.join(","));
+console.log("inspect", [true, false], g);
+
+// Sets of booleans in both directions.
+const fromArray = new Set<boolean>([true, true, false, true]);
+console.log("set", fromArray.size, fromArray.has(false));
+const back: boolean[] = [...fromArray];
+back.push(false);
+show("set-spread", back);
+show("set-from", Array.from(new Set<boolean>([false, true])));
+
+// Nested arrays and flat.
+const grid: boolean[][] = [[true, false], [false], [true, true, false]];
+grid[1]![3] = true;
+show("flat", grid.flat());
+console.log("grid", grid.map((row) => row.join("")).join("/"));
+
+// Present undefined stays distinct from a hole.
+const m: (boolean | undefined)[] = [true, undefined, false];
+m[4] = true;
+showMaybe("maybe", m);
+m.splice(1, 0, undefined, false);
+showMaybe("maybe-splice", m);
+showMaybe("maybe-slice", m.slice(1));
+showMaybe("maybe-reverse", m.reverse());
+console.log("maybe-includes", m.includes(undefined), m.indexOf(undefined), m.indexOf(false));
+console.log("maybe-json", JSON.stringify(m));
+const n: (boolean | undefined)[] = new Array<boolean | undefined>(4).fill(undefined);
+n[2] = true;
+showMaybe("maybe-fill", n);
+
+// Growth across allocation sizes, large sparse indices and densifying.
+const big: boolean[] = [];
+for (let i = 0; i < 70000; i++) big.push(i % 3 === 0);
+let bigTrues = 0;
+for (let i = 0; i < big.length; i++) if (big[i]) bigTrues++;
+console.log("big", big.length, bigTrues, big[69999], big.lastIndexOf(true));
+const sparse: boolean[] = [true];
+sparse[3000000] = true;
+sparse[2000000] = false;
+console.log("sparse", sparse.length, 1 in sparse, sparse[2000000], sparse[3000000], sparse.indexOf(false));
+for (let i = 1; i < 200; i++) sparse[i] = i % 2 === 0;
+console.log("sparse-dense", sparse.slice(0, 6).join(","), sparse.indexOf(false), sparse.includes(false));
+sparse.length = 4;
+show("sparse-truncate", sparse);
+const sieve: boolean[] = new Array<boolean>(100001).fill(false);
+let primes = 0;
+for (let i = 2; i <= 100000; i++) {
+  if (sieve[i]) continue;
+  primes++;
+  for (let j = i * i; j <= 100000; j += i) sieve[j] = true;
+}
+console.log("sieve", primes, sieve.indexOf(false, 100), sieve.lastIndexOf(false));
diff --git a/tests/corpus/5081-boolean-array-dynamic-views.ts b/tests/corpus/5081-boolean-array-dynamic-views.ts
new file mode 100644
index 000000000..8b73b805a
--- /dev/null
+++ b/tests/corpus/5081-boolean-array-dynamic-views.ts
@@ -0,0 +1,23 @@
+// A boolean[] seen through unknown and any: reads, writes and growth through
+// the dynamic view land in the same one-byte element storage, and JSON,
+// structuredClone, deep equality and inspect read it back.
+import { deepStrictEqual } from "node:assert";
+import { inspect, isDeepStrictEqual } from "node:util";
+const flags: boolean[] = [true, false, true];
+flags[5] = false;
+const view: unknown = flags;
+console.log(Array.isArray(view), JSON.stringify(view));
+const anyView = view as any;
+console.log(anyView[0], anyView[1], anyView[3], anyView.length);
+anyView[1] = true;
+anyView[3] = true;
+console.log(flags[1], flags[3], 4 in flags, flags.length);
+const cloned = structuredClone(flags);
+console.log(cloned.length, cloned.join(","), isDeepStrictEqual(cloned, flags));
+deepStrictEqual([true, false], [true, false]);
+console.log(inspect(flags), inspect([false]));
+const back = view as boolean[];
+back.push(true);
+console.log(flags.length, flags[6]);
+const rec: Record<string, unknown> = { flags };
+console.log(JSON.stringify(rec));
diff --git a/tests/harness/self-hosting-native-toolchain.test.ts b/tests/harness/self-hosting-native-toolchain.test.ts
index 2906ac1e2..17665e888 100644
--- a/tests/harness/self-hosting-native-toolchain.test.ts
+++ b/tests/harness/self-hosting-native-toolchain.test.ts
@@ -96,7 +96,7 @@ for (const backend of ["llvm"] as const) {
           object_format: target.objectFormat,
           minimum_os: target.minimumOs,
         },
-        runtime_abi: { version: 17, marker: "scr_runtime_abi_v17" },
+        runtime_abi: { version: 18, marker: "scr_runtime_abi_v18" },
         compiler: { command: "fixture", identity: "fixture", target: target.llvmTriple },
         macros: { executable: [], excluded: [], sanitizer: "external-toolchain-required" },
         flavors: {