Zig Version

master

Steps to Reproduce and Observed Behavior

Create a lib.c file with following content:

typedef int i32x2 __attribute__((__vector_size__(8)));
typedef float f32x2 __attribute__((__vector_size__(8)));

f32x2 cast_function(i32x2 x) {
  return (f32x2) __builtin_convertvector((i32x2) { x[0], x[1] }, f32x2);
}

Run zig translate-c lib.c. The function will transform into the following zig code:

pub export fn cast_function(arg_x: i32x2) f32x2 {
    var x = arg_x;
    _ = &x;
    return blk: {
        const tmp = @as(f32, @floatFromInt((blk_1: {
            const tmp = x[@as(c_uint, @intCast(@as(c_int, 0)))];
            const tmp_2 = x[@as(c_uint, @intCast(@as(c_int, 1)))];
            break :blk_1 i32x2{
                tmp,
                tmp_2,
            };
        })[0]));
        const tmp_1 = @as(f32, @floatFromInt((blk_1: {
            const tmp = x[@as(c_uint, @intCast(@as(c_int, 0)))];
            const tmp_2 = x[@as(c_uint, @intCast(@as(c_int, 1)))];
            break :blk_1 i32x2{
                tmp,
                tmp_2,
            };
        })[1]));
        break :blk f32x2{
            tmp,
            tmp_1,
        };
    };
}

When trying to compile this code (either directly, or by just c-importing it into zig code) we get an error:

/.../cimport.zig:309:19: error: local constant 'tmp' shadows local constant from outer scope
            const tmp = x[@as(c_uint, @intCast(@as(c_int, 0)))];
                  ^~~
/.../cimport.zig:300:15: note: previous declaration here
        const tmp = @as(f32, @floatFromInt((blk_1: {

Expected Behavior

The C code should be translated without naming collisions. Apparently something is not fully working with how compiler generates unique temporary names.

For reference: https://clang.llvm.org/docs/LanguageExtensions.html#builtin-convertvector

I am also a little bit surprised by another thing. Part const tmp = is mentioned three times in generated code. But the error is produce for collision between first and third mentions. Why then the second mention is not shadowing the first one.

Question: I wrote a simple test in test/cases/translate_c/ to reproduce the issue (with faulty zig code output). Should I create a pull request so once the issue is fixed we can alter the test and ensure that the output is good now?

Submitting a bug here to ensure this is not a known issue.