save_image unsupported operation

[Timmmm]

Maybe I'm using it wrong (thought it's hard to tell because there's not really any documentation), but save_image() always gives me this error:

thread 'main' panicked at 'Error message: Unsupported operation/parameter option', C:\Users\Tim\.cargo\registry\src\.com-1ecc6299db9ec823\arrayfire-3.4.2\src\error.rs:66

I've tried PNG, JPEG, f32 arrays, u8, 3D, 2D, etc. It really needs better error reporting too.

[9prady9]

Can please share the code stub that can help me reproduce the issue. Also, please do export AF_PRINT_ERRORS=1 from command line and then run the program to see a more detailed error message.

[Timmmm]

Here's a fairly minimal test case:

extern crate arrayfire as af;

use af::*;

fn main() {
	set_device(0);
	info();

	let win = Window::new(100, 100, "Test".to_string());

	let values = [0u8, 0];
	let mat = Array::new(&values, Dim4::new(&[2, 1, 1, 1]));

	let mut it: i32 = 0;
	while !win.is_closed() {

		win.draw_image(&mat, None);
		save_image(format!("image_{}.png", it), &mat);

		it += 1;
		if it > 20 {
			break;
		}
	}
}

What's weird is that sometimes it creates between 1 and 5 images and then fails. Here's the output:

ArrayFire v3.4.1 (CUDA, 64-bit Windows, build b9055b1)
Platform: CUDA Toolkit 8, Driver: CUDA Driver Version: 8000
[0] GeForce GTX 970, 4096 MB, CUDA Compute 5.2
In function af_err __cdecl af_save_image(const char *,void *const )
In file src\api\c\imageio.cpp:297
FreeImage Error: Unknown Filetype

thread 'main' panicked at 'Error message: Unsupported operation/parameter option', C:\Users\Tim\.cargo\registry\src\.com-1ecc6299db9ec823\arrayfire-3.4.2\src\error.rs:66
note: Run with `RUST_BACKTRACE=1` for a backtrace.
CUDA Error (30): unknown error

That makes it seem like it doesn't support PNG, but it has actually created valid PNG files occasionally so that can't be true. Anyway, trying .jpg gives the same error, but again it actually creates a few jpegs. It seems to reliably create 3 for me, and they seem to be valid.

[Timmmm]

I wonder if this line is safe:

    let err_val = af_save_image(filename.clone().as_bytes().as_ptr() as *const u8,

Is there a risk the temporary created by .clone() is destroyed before af_save_image() is called?

Edit: Nah that doesn't seem to be it.

[9prady9]

Good catch. While we corrected this issue in all the other modules, somehow image module was not rectified. The root cause is that when we transfer a string to C-FFI, we need to build a CString from String instead of clone and then pass it onto FFI. I fixed this in devel now.

[Timmmm]

Awesome, works like a charm. I still can't work out exactly what to save_image() so I don't just get a black image but save_image_native("foo.tif", ...) with an f32 array worked perfectly.

Can you explain to an unlearned Rustacean what exactly was wrong with the original code?

[9prady9]

There is a difference in how rust represents the strings using std::ffi::Cstring and std::string::String.

std::ffi::CString

A type representing an owned C-compatible string

This type serves the primary purpose of being able to safely generate a
C-compatible string from a Rust byte slice or vector. An instance of this
type is a static guarantee that the underlying bytes contain no interior 0
bytes and the final byte is 0.

where as

std::string::String

A UTF-8 encoded, growable string and it stores the underlying data in the
form of vec<u8>. Also, there is no guarantee of having the final 0-byte that
is expected by the C `const char*` strings.

This was the issue we fixed.