Replace uninitialized with MaybeUninit

[Demi-Marie]

mem::uninitialized is deprecated and unsafe. This replaces its use
with mem::MaybeUninit and adds a proof that the use of
mem::MaybeUninit is correct.

Requested by @niklasad1.

Fixes #238.

[parity-cla-bot]

It looks like @demimarie-parity hasn't signed our Contributor License Agreement, yet.

The purpose of a CLA is to ensure that the guardian of a project's outputs has the necessary ownership or grants of rights over all contributions to allow them to distribute under the chosen licence.
Wikipedia

You can read and sign our full Contributor License Agreement at the following URL: https://cla.parity.io

Once you've signed, please reply to this thread with [clabot:check] to prove it.

Many thanks,

Parity Technologies CLA Bot

[ordian]

@demimarie-parity have you tried to measure if zeroing memory has no perf impact? (#233 (comment))

[Demi-Marie]

@ordian I haven’t. It would be worthwhile, though ― LLVM might even be able to optimize out the initialization.

[niklasad1]

@ordian
I checked it ended up with identical assembly

[Demi-Marie]

@niklasad1 how did you do this? Also @ordian my latest commit removes all use of uninitialized memory.

[niklasad1]

I created temporary initialization of U1024 in src/lib.rs

then this:

diff --git a/uint/src/lib.rs b/uint/src/lib.rs
index 352ccf7..751766d 100644
--- a/uint/src/lib.rs
+++ b/uint/src/lib.rs
@@ -31,3 +31,8 @@ pub use crunchy::unroll;
 #[macro_use]
 mod uint;
 pub use crate::uint::*;
+
+
+construct_uint! {
+       pub struct U1024(16);
+}
diff --git a/uint/src/uint.rs b/uint/src/uint.rs
index cc58ee2..0cdec03 100644
--- a/uint/src/uint.rs
+++ b/uint/src/uint.rs
@@ -72,13 +72,14 @@ macro_rules! impl_try_from_for_primitive {
 
 #[macro_export]
 #[doc(hidden)]
+#[inline(never)]
 macro_rules! uint_overflowing_binop {
        ($name:ident, $n_words: tt, $self_expr: expr, $other: expr, $fn:expr) => ({
                let $name(ref me) = $self_expr;
                let $name(ref you) = $other;
 
-               let mut ret = unsafe { $crate::core_::mem::uninitialized() };
-               let ret_ptr = &mut ret as *mut [u64; $n_words] as *mut u64;
+               let mut ret = [0_u64; $n_words];
+               let ret_ptr = ret.as_mut_ptr();
                let mut carry = 0u64;
 
                unroll! {
@@ -876,7 +877,7 @@ macro_rules! construct_uint {
                        }
 
                        /// Add with overflow.
-                       #[inline(always)]
+                       #[inline(never)]
                        pub fn overflowing_add(self, other: $name) -> ($name, bool) {
                                uint_overflowing_binop!(
                                        $name,
(END)

finally,

$ cargo asm --no-color uint::U1024::overflowing_add

[niklasad1]

👍

BTW, is isize::MAX used because of offset?

[niklasad1]

Suggested change

	let ret_ptr = &mut ret as *mut [u64; $n_words] as *mut u64;
	let ret_ptr = ret.as_mut_ptr();

[niklasad1]

LGTM

[ordian]

The zero initialization and safety comments look good to me, there is a small caveat with static_assertions bump, but I think it's fine.

[ordian]

I'm afraid this is technically a breaking change, since we pub use static_assertions;, but this is unlikely to be a problem in practice.

[ordian]

needs resolving

[dvdplm]

needs resolving

Can you please clarify what it is that needs resolving here?

[ordian]

Can you please clarify what it is that needs resolving here?

Sorry, I meant merge conflict needs to be resolved.