About the mistranslation

[King-play]

Original C Code:

int main(void) {
    assert(check_add());    // once 
    assert(check_sub());    // once
    return 0;
}

C2rust Output:

unsafe fn main_0() -> core::ffi::c_int {
    // first
    if check_add() {} else {
        __assert_fail(...);
    }
    
    // 🔴 second !!!
    'c_1569: {
        if check_add() {} else {
            __assert_fail(...);
        }
    };
    
    // first
    if check_sub() {} else {
        __assert_fail(...);
    }
    
    // 🔴 second !!!
    'c_1529: {
        if check_sub() {} else {
            __assert_fail(...);
        }
    };
    
    return 0 as core::ffi::c_int;
}

[kkysen]

Weird. Thanks for finding this. Full reproduction on godbolt: https://godbolt.org/z/E5W37fn5E

#include <assert.h>
#include <stdbool.h>

bool check_add(void);
bool check_sub(void);

int main(void) {
    assert(check_add());    // once 
    assert(check_sub());    // once
    return 0;
}

[kkysen]

Oh, this is just a more minimal example of #1507, right?

[King-play]

Yes. I wish this bug can be sloved, thank you!

[Rua]

It looks like the code posted in #1514 (comment) is also running into the same issue. Using the macro-expanded code provided in that issue I was able to narrow down the source. The following triggers the bug:

(
     (void) sizeof ((jso->_ref_count < (4294967295U)) ? 1 : 0),
     __extension__ ({
          if (jso->_ref_count < (4294967295U)) ;
          else __assert_fail (
               "jso->_ref_count < UINT32_MAX",
               "json_object.c",
               174,
               __extension__ __PRETTY_FUNCTION__
          );
     });
);

but this does not:

     __extension__ ({
          if (jso->_ref_count < (4294967295U)) ;
          else __assert_fail (
               "jso->_ref_count < UINT32_MAX",
               "json_object.c",
               174,
               __extension__ __PRETTY_FUNCTION__
          );
     });

So it looks like the culprit may in the translation of statement expressions or the dreaded comma operator.

[Rua]

I'm almost certain the problem lies in this match block somewhere:

c2rust/c2rust-transpile/src/translator/mod.rs

Lines 4375 to 4397 in c23cba2

	match as_semi_break_stmt(&stmt, &lbl) {
	Some(val) => {
	let block = mk().block_expr(match val {
	None => mk().block(stmts),
	Some(val) => WithStmts::new(stmts, val).to_block(),
	});
	// enclose block in parentheses to work around
	// https://github.com/rust-lang/rust/issues/54482
	let val = mk().paren_expr(block);
	let stmts = if ctx.is_unused() {
	vec![mk().semi_stmt(val.clone())]
	} else {
	Vec::new()
	};
	return Ok(WithStmts::new(stmts, val));
	}
	_ => {
	self.use_feature("label_break_value");
	stmts.push(stmt)
	}
	}

Before that block, the stmts variable has a value that looks correct, containing only a single statement for the if. Since the label_break_value feature is being emitted, that must mean it's taking the _ branch of the match, which must mean that as_semi_break_stmt is returning None. I don't yet understand the logic of what this code is doing with labels beyond here.