RFC: allow some intrinsics in static eval context

active/0000-static-intrinsics.md

@@ -0,0 +1,51 @@
+- Start Date: 2014-04-29
+- RFC PR #: (leave this empty)
+- Rust Issue #: (leave this empty)
+
+# Summary
+
+Allow the use of some intrinsics in static evaluation context, specifically
+`init`, `uninit` and `size_of`.
+
+# Motivation
+
+Intrinsics provide valuable details or initialization for statics, that are much
+more widely used in embedded. While all the operations are unsafe by nature,
+the careful use by programmer should be allowed.
+
+# Detailed design
+
+Most of the use cases are related to instantiation of a struct in a pre-defined
+memory location. This might or might not be solved by any custom allocators.
+
+Example 1: initialize an object in `.bss`:
+
+```rust
+static mut inst: T = unsafe { uninit() };
+```
+
+this would allocate the required chunk in .bss, the access to the `inst` is
+undefined, until it's properly initialized.
+
+Example 2: initialize a buffer for objects:
+
+```rust
+static mut inst_buf: [u8, ..size_of::<T>() * 4] = [0, ..size_of::<T>() * 4];
+```
+
+this would allocate the chuck sufficient to hold 4 `T`s in .bss. This case
+doesn't solve the instantiation, though, `memcpy` might be used for copying an
+object from stack to `inst_buf`, but that's a really unsafe option.
+
+
+# Alternatives
+
+Provide more management for `~T`s, so that the actual heap / location in heap
+could be specified. Anything that might be allocated in heap must be strictly
+verified on case by case basis.
+
+# Unresolved questions
+
+Is it `static mut inst_buf: [u8, ..size_of::<T>() * 4]` or
+`static mut inst_buf: [T, ..4]`? If we can use `init`/`uninit`, the second
+option is acceptable.