libpng: use of undefined value error with png_check_sig

[naeu]

.../cimport.zig:26019:100: error: use of undefined value here causes undefined behavior
pub inline fn png_check_sig(sig: anytype, n: anytype) @TypeOf(!(png_sig_cmp(sig, @as(c_int, 0), n) != 0)) {
                                                                                                   ^

Test case:

// file: main.zig
// zig build-exe main.zig -lpng -lc
const c = @cImport({
    @cInclude("png.h");
});
pub fn main() void {
    var buf: [100]u8 = undefined;
    _ = c.png_check_sig(buf[0..], 100);
}

Workaround:

fn png_check_sig(sig: [*]const u8, n: usize) c_int {
    return @boolToInt(c.png_sig_cmp(sig, 0, n) == 0);
}

zig version: 0.9.0-dev.959+f011f1393

[sskras]

If I get the error message right, the inline form is functionally equal to this:

fn png_check_sig(sig: anytype, n: anytype)  c_int {
    return !         (c.png_sig_cmp(sig, 0, n) != 0);
}

If we run unified diff between it and the workardound, the difference would be:

-    return !         (c.png_sig_cmp(sig, 0, n) != 0);
+    return @boolToInt(c.png_sig_cmp(sig, 0, n) == 0);

By looking at this and the label:"contributor friendly", as a potential contributor I wonder:
what needs to be done and would it be appropriate for a newbie?

Maybe the @boolToInt should just be inserted between the ! and the first ( in the original code, so it becomes:

    return !@boolToInt(c.png_sig_cmp(sig, 0, n) != 0);

?

[sskras]

Maybe the @boolToInt should just be inserted between the ! and the first ( in the original code, so it becomes:

    return !@boolToInt(c.png_sig_cmp(sig, 0, n) != 0);

I see I got this very much wrong:

1. ! will not work with integers;
2. Even if something is to be "inserted" here, I meant doing this not figuratively but logically, probably on the parser level.

---

OK, back to the source:

• C version, the originals:

int png_sig_cmp(png_bytep sig, png_size_t start, png_size_t num_to_check);

#define png_check_sig(sig, n) !png_sig_cmp((sig), 0, (n))

The original prototype of png_check_sig(), now obsolete and converted into the aforementioned macro:

int png_check_sig(png_bytep arg0, int arg1);

It looks like both names should return int.

• Zig version, autotranslated:

pub fn png_sig_cmp(sig: png_const_bytep, start: usize, num_to_check: usize) c_int;

pub inline fn png_check_sig(sig: anytype, n: anytype) @TypeOf(!(png_sig_cmp(sig, @as(c_int, 0), n) != 0)) {
    return !(png_sig_cmp(sig, @as(c_int, 0), n) != 0);
}

---

I think the original error message can be reduced to:

.../cimport.zig:26019:100: error: use of undefined value here causes undefined behavior

@TypeOf(!(png_sig_cmp(sig, @as(c_int, 0), n) != 0))
                                             ^

Now if I try to print the @typeName separately and use a constants in place of sig and n, Zig says the type is bool instead of c_int.

Still not sure what type should be returned by the translated function.

[david-vanderson]

Looks like this issue got fixed somewhere between 0.10.0-dev.2431+0e6285c8f and 0.10.0-dev.4060+61aaef0b0 (which just happen to be two versions I have lying around)

[tau-dev]

So can this be closed?

[Vexu]

It can be closed when a test is added.

[Vexu]

The reduced version of this issue is trivial and doesn't reproduce on stage2 due to properly implementing #1947

test {
    const a: bool = undefined;
    _ = !a;
}