Limit to write unused unary expr only when expr is not pure

[bungcip]

currently, c2rust transpiled this code

void increment(){
    int i = 5;
    i++;
    i--;
    --i;
    ++i;
}

to this

#[no_mangle]
pub unsafe extern "C" fn increment() {
    let mut i: std::ffi::c_int = 5 as std::ffi::c_int;
    i += 1;
    i;
    i -= 1;
    i;
    i -= 1;
    i;
    i += 1;
    i;
}

which contain useless i expression. this is because UnaryExpr can contain side effect expression which cannot be ignored, however not all unary expr can contain side effects. this pull request limit to only Negate , Plus, Complement, and Not . However we can limit it to only expr which is not pure.

[kkysen]

For the CI failure, if you're only developing on Linux, the macOS-specific snapshots won't be automatically updated so you have to manually update them based on what CI (or a macOS machine) says.

[kkysen]

Oh, actually, this test is OS specific since the printf declarations are slightly different. It's an annoying difference, so you could just try to rewrite it without using printf for the side effect (or just declaring it yourself).

[bungcip]

ok, now testcase is not using printf. let see the this PR pass the testsuite.

[kkysen]

Thanks for the help on this!

[kkysen]

Suggested change

	// to add them to stmts when arg expr is not pure
	// to add them to stmts when `arg` is not pure.

[kkysen]

So these unary ops now emit nothing, and while that's generally correct (I haven't thought through the exact semantics of the *& one yet), it may also delete code that was intentionally written in C, similar to #1326. For things like i++, removing the i; and only keeping the i += 1; makes perfect sense, but I'm not sure for these other unary operators.

[bungcip]

hmm, i thought that c2rust have policy to removing unused things, i saw some other code that skip generating rust ouput when ctx.is_unused() is true, if this is wrong I will the change the code to remove in increment expression only.