Observing what looks like suboptimal codegen when loading and assigning Bool variables. It seems to undergo some transformation where I'm guessing it loses the information that it's a Bool type and upon assignment forces its value to [0, 1] by binary and, which at least naively to me seems redundant.

Example below, tested on 0.5.1 and 0.6.0-pre.alpha.

mutable struct Foo
    flag1::Bool
    flag2::Bool
end

function update(foo::Foo)
    foo.flag2 = foo.flag1
end

code_native(update, (Foo,))
code_llvm(update, (Foo,))

LLVM output

define i8 @julia_update_60814(i8** dereferenceable(2)) #0 !dbg !5 {
top:
  %1 = bitcast i8** %0 to i8*
  %2 = load i8, i8* %1, align 16
  %3 = getelementptr i8, i8* %1, i64 1
  %4 = and i8 %2, 1
  store i8 %4, i8* %3, align 1
  ret i8 %4
}

Native x64 output

	.text
Filename: In[1]
	pushq	%rbp
	movq	%rsp, %rbp
Source line: 9
	movb	(%rdi), %al
	andb	$1, %al
	movb	%al, 1(%rdi)
	popq	%rbp
	retq
	nopl	(%rax)