RFC: zero initializer / undefined intrinsics for POD types

[nikomatsakis]

Perhaps we should add an intrinsic like:

fn zero<T: POD>() -> T

that basically fills T with zeros. This is perfectly safe for POD types. Of course, we don't have a POD kind yet, but that's a separate issue I guess. This is particularly useful in C interop, where you often have functions with out-params that otherwise need to be initialized by hand.

We could also add an intrinsic like

fn undefined<T: POD>() -> T

which would just leave the data uninitialized. This would be more efficient but less safe in some sense. Though restricting it to POD at least preserves memory safety, more-or-less.

The idea is you'd write something like:

struct FooResult { /*something POD*/ }

extern mod c_library { fn foo(p: &mut FooResult) -> bool; }

fn foo() -> FooResult {
    let result = defined();
    if !c_library::foo(&mut result) { fail; }
    return result;
}

[nikomatsakis]

Here is an example that came up on IRC, but it comes up regularly: http://pastebin.com/v534PrDW. Sscroll to the final function.

[jruderman]

Though restricting undefined to POD at least preserves memory safety, more-or-less.

In C++, POD structs can contain pointers.

[jruderman]

SpiderMonkey contains a similar PodZero function:

https://bugzilla.mozilla.org/show_bug.cgi?id=551276
http://hg.mozilla.org/mozilla-central/annotate/87f6f50a0207/js/src/jsutil.h#l177

[jruderman]

I'd prefer to declare the C function as taking a &write FooResult, so the type-checker knows it can be uninitialized before the call and will be initialized after the call.

[nikomatsakis]

In C++, POD structs can contain pointers.

Yeah, I wrote "more-or-less" because I figured *T would be POD, but not @T or ~T, but of course we don't make any guarantees about unsafe ptrs.

[nikomatsakis]

So, it turns out that this exists! (the intrinsic init()) However, it does not check for POD and nor is it considered unsafe. Bad. See https://github.com/mozilla/rust/blob/master/src/libcore/comm.rs#L297. I'm gonna repurpose this issue to address its shortcomings.

[nikomatsakis]

Oh, never mind, init() seems to fill with zero. It should still be unsafe.

[Aatch]

Closed by #6354