Use copy_nonoverlapping instead of slice::copy_from

[jswrenn]

In doing so, we eliminate a potential panic path. Although I was not able to observe panic paths emitted in toy examples, they might be emitted in complex examples in which the optimizer is low on gas. Regardless, I expect this change will ease our future adoption of call-graph analysis techniques of potential panic paths.

Ref #200 (comment)

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Please upload report for BASE (main@e12c1e9). Learn more about missing BASE report.
Report is 7 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1448   +/-   ##
=======================================
  Coverage        ?   87.45%           
=======================================
  Files           ?       15           
  Lines           ?     5166           
  Branches        ?        0           
=======================================
  Hits            ?     4518           
  Misses          ?      648           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[joshlf]

Suggested change

	if bytes.len() == mem::size_of_val(self) {
	// SAFETY: Within this branch of the conditional, we have ensured
	// that `bytes.len()` (the size of the destination) is equal to
	// `mem::size_of_val(self)` the size of the source value. Neither
	// the size of the source nor the size of the destination change
	// between the above size check and the invocation of
	// `copy_unchecked`. The size of `self` (the source) cannot change,
	// because `Self` is bounded as `Immutable`, and the size of `bytes`
	// cannot change, because we hold an exclusive reference to it and
	// do not change its size.
	unsafe { util::copy_unchecked(self.as_bytes(), bytes) }
	let self_bytes = self.as_bytes();
	if bytes.len() == self_bytes.len() {
	// SAFETY: Within this branch of the conditional, we have ensured
	// that `bytes.len()` (the size of the destination) is equal to
	// `self_bytes.len()` (the size of the source). Neither
	// the size of the source nor the size of the destination change
	// between the above size check and the invocation of
	// `copy_unchecked`.
	unsafe { util::copy_unchecked(self_bytes, bytes) }

(And reflow the safety commnet)

This makes the safety argument simpler since we don't need to reason about the relationship between size_of_val(self) and self.as_bytes(). It is probably ever so slightly tougher on the optimizer, but I doubt it's enough to make a difference.

[jswrenn]

Done

[joshlf]

Same here as above - do self.as_bytes() first and bounds check using self_bytes.len().

[jswrenn]

Done

[joshlf]

Same here as above - do self.as_bytes() first and bounds check using self_bytes.len().

[jswrenn]

Done

[joshlf]

Suggested change

	// - `dst.as_ptr()` is valid for writes of `dst.len.len()` bytes, because the
	// - `dst.as_ptr()` is valid for writes of `src.len()` bytes, because the

Typo?

[jswrenn]

Done