Implement ResizableArrayBuffer [wip]

[littledivy]

https://github.com/tc39/proposal-resizablearraybuffer

Changes:

• Add max_byte_length and resizable fields to JSArrayBuffer
• Add is_length_tracking and is_backed_by_rab fields to JSTypedArray
• Add js_array_buffer_constructor4 for constructing RAB with maxByteLength
• Implement resize() on top of js_realloc
• Handle variable length tracking for JSTypedArray

Did not implement SharedArrayBuffer in this patch.

[bnoordhuis]

Run build/run-test262 -c test.conf -a -u to update test262_errors.txt.

[bnoordhuis]

This leaks val (when it's not an int32), JS_ToIndex doesn't free it.

[bnoordhuis]

I realize the code isn't 100% consistent in this respect but let's drop the braces for single-line if statements.

[bnoordhuis]

Can be just:

Suggested change

	if (abuf->detached) {
	JS_ThrowTypeErrorDetachedArrayBuffer(ctx);
	return JS_EXCEPTION;
	}
	if (abuf->detached)
	return JS_ThrowTypeErrorDetachedArrayBuffer(ctx);

[bnoordhuis]

Should handle reallocation failures. The new_len == 0 case is kind of tricky because the allocator can either return NULL or a unique pointer. Maybe easiest to js_free() when new_len == 0.

[bnoordhuis]

Suggested change

	return JS_NewInt32(ctx, (abuf->byte_length - ta->offset));
	return JS_NewUint32(ctx, abuf->byte_length - ta->offset);

Because ta->offset is uint32_t. abuf->byte_length probably should be too but one thing at a time.

[littledivy]

It doesn't look like we can use realloc for this:

Resiazable AB length can shrink mid iteration (eg: Array#every) and the tests expect the initial buffer elements to not be freed. So we want the backing buffer to not move and implement this by reserving pages and using them during resize (like V8)

[bnoordhuis]

Resiazable AB length can shrink mid iteration (eg: Array#every) and the tests expect the initial buffer elements to not be freed. So we want the backing buffer to not move and implement this by reserving pages and using them during resize (like V8)

If you mean mmap'ing memory, I don't think we want to go there just yet. Can you point me to a test that triggers that behavior?

[littledivy]

Can you point me to a test that triggers that behavior?

https://github.com/tc39/test262/blob/c1281dba453b4d97f2bc04b8dfe722c5d5f90b89/test/built-ins/Array/prototype/every/callbackfn-resize-arraybuffer.js

[littledivy]

I guess I'll close this one for now until we wanna mmap

[bnoordhuis]

For posterity: RABs landed just now via #646 - and without an mmap call in sight :-)