derive(Debug) on struct containing array that is too big for derive(Debug)

[fitzgen]

$ ./target/debug/bindgen --enable-cxx-namespaces test.hpp > test.rs 

Here is test.hpp (really really nasty):

template <typename a, int b> class c { a e[b]; };
class d;
template <typename g, g f> class C { c<d, f> h; };
class i {
  i *j;
  i *k;
  bool l;
};
class d {
  i m;
};
enum n { o, p, q, r, s, t, b, ae, e, ag, ah, ai };
class F {
  C<n, ai> w;
};

Here is the generated test.rs bindings:

/* automatically generated by rust-bindgen */

pub mod root {
    #[allow(unused_imports)]
    use self::super::root;
    #[repr(C)]
    #[derive(Debug, Copy)]
    pub struct d {
        pub m: root::i,
    }
    #[test]
    fn bindgen_test_layout_d() {
        assert_eq!(::std::mem::size_of::<d>() , 24usize);
        assert_eq!(::std::mem::align_of::<d>() , 8usize);
    }
    impl Clone for d {
        fn clone(&self) -> Self { *self }
    }
    #[repr(C)]
    #[derive(Debug, Copy)]
    pub struct i {
        pub j: *mut root::i,
        pub k: *mut root::i,
        pub l: bool,
    }
    #[test]
    fn bindgen_test_layout_i() {
        assert_eq!(::std::mem::size_of::<i>() , 24usize);
        assert_eq!(::std::mem::align_of::<i>() , 8usize);
    }
    impl Clone for i {
        fn clone(&self) -> Self { *self }
    }
    #[repr(u32)]
    #[derive(Debug, Copy, Clone, PartialEq, Eq, Hash)]
    pub enum n {
        o = 0,
        p = 1,
        q = 2,
        r = 3,
        s = 4,
        t = 5,
        b = 6,
        ae = 7,
        e = 8,
        ag = 9,
        ah = 10,
        ai = 11,
    }
    #[repr(C)]
    #[derive(Debug, Copy)]
    pub struct F {
        pub w: [u64; 33usize],
    }
    #[test]
    fn bindgen_test_layout_F() {
        assert_eq!(::std::mem::size_of::<F>() , 264usize);
        assert_eq!(::std::mem::align_of::<F>() , 8usize);
    }
    impl Clone for F {
        fn clone(&self) -> Self { *self }
    }
}

And when we try to compile the bindings, we get this error:

$ rustc --crate-type lib test.rs 
error[E0277]: the trait bound `[u64; 33]: std::fmt::Debug` is not satisfied
  --> js.rs:53:9
   |
53 |         pub w: [u64; 33usize],
   |         ^^^^^^^^^^^^^^^^^^^^^ the trait `std::fmt::Debug` is not implemented for `[u64; 33]`
   |
   = note: `[u64; 33]` cannot be formatted using `:?`; if it is defined in your crate, add `#[derive(Debug)]` or manually implement it
   = note: required because of the requirements on the impl of `std::fmt::Debug` for `&[u64; 33]`
   = note: required for the cast to the object type `std::fmt::Debug`

error: aborting due to previous error

[emilio]

This is a nasty consequence of us trying to do the best we can to guarantee correct layout.

When we find a type with non-type template parameters, we generate a blob of memory. Of course, we don't check that with can_derive_debug.

We should. We can go safe and just return false if the type had any non-type template parameters, or we can factor out the logic from BlobTyBuilder, and check the layout of the type to see if it'd generate a valid array.

We should probably do the same for can_derive_copy.

[emilio]

@fitzgen do yo want to tackle or mentor this bug?

[fitzgen]

I can take it.

[64]

I'm still getting this error while trying to run bindgen on uv.h:

error[E0277]: the trait bound `[u8; 40]: std::fmt::Debug` is not satisfied
     --> c:\code\rust\libuv-test\target\debug\build\libuv-test-9b56431e6ab39041\out/bindings.rs:75085:5
      |
75085 |     pub __bindgen_padding_0: [u8; 40usize],
      |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `std::fmt::Debug` is not implemented for `[u8; 40]`
      |
      = note: `[u8; 40]` cannot be formatted using `:?`; if it is defined in your crate, add `#[derive(Debug)]` or manually implement it
      = note: required because of the requirements on the impl of `std::fmt::Debug` for `&[u8; 40]`
      = note: required for the cast to the object type `std::fmt::Debug`

error: aborting due to previous error

I'm using bindgen version 0.26.2 on windows. Any ideas?

[emilio]

That's a different issue, and it deserves another bug, I think. That's when generating padding for a struct, which is a different case.

I can file as soon as I arrive home.

[tomwhoiscontrary]

I'm getting some relative of this. I am translating some reasonably big C++ class, and i tried making it opaque to make my life easier:

#[repr(C)]
#[derive(Debug)]
pub struct OsiClpSolverInterface {
    pub _bindgen_opaque_blob: [u64; 112usize],
}

But this fails to compile:

error[E0277]: the trait bound `[u64; 112]: std::fmt::Debug` is not satisfied
    --> /home/tomwhoiscontrary/src/github/initech/clp-sys/target/debug/build/clp-sys-515894cd11841cff/out/bindings.rs:2993:5
     |
2993 |     pub _bindgen_opaque_blob: [u64; 112usize],
     |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `std::fmt::Debug` is not implemented for `[u64; 112]`
     |

If i make the type non-opaque, it comes out like this:

#[repr(C)]
#[derive(Debug)]
pub struct OsiClpSolverInterface {
    pub vtable_: *const OsiClpSolverInterface__bindgen_vtable,
    // skip numerous fields
    pub __bindgen_padding_0: [u64; 37usize],

Which also fails:

error[E0277]: the trait bound `[u64; 37]: std::fmt::Debug` is not satisfied
    --> /home/tomwhoiscontrary/src/github/initech/clp-sys/target/debug/build/clp-sys-515894cd11841cff/out/bindings.rs:3122:5
     |
3122 |     pub __bindgen_padding_0: [u64; 37usize],
     |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `std::fmt::Debug` is not implemented for `[u64; 37]`
     |

The latter case looks similar to @64's bug. I wonder if this is related to #648?

I can work around this by setting .derive_debug(false). I don't think the debugging printout on these objects is going to be very helpful anyway!

As an aside, should there really be 37 64-bit words of padding? I don't know much about C++ object layout at all, but that surprised me.

[fitzgen]

@tomwhoiscontrary would you mind filing a new issue and including a test case so we can reproduce the issue locally? Thanks!

[tomwhoiscontrary]

I'll see what i can do about a test case. The code i'm binding is pretty huge, so i'll need to extract a subset which exhibits the problem.

[fitzgen]

The code i'm binding is pretty huge, so i'll need to extract a subset which exhibits the problem.

This is where C-Reduce and bindgen::Builder::dump_preprocessed_input come in really handy ;)

https://github.com/rust-lang-nursery/rust-bindgen/blob/master/CONTRIBUTING.md#using-creduce-to-minimize-test-cases

[mmacedoeu]

That's a different issue, and it deserves another bug, I think. That's when generating padding for a struct, which is a different case.

I can file as soon as I arrive home.

@emilio
I got that bug today running on windows as well, did you manage to file the bug and maybe there is a fix already ?

[martinthomson]

FYI, this is what I'm getting after upgrading to 1.39:

error[E0277]: arrays only have std trait implementations for lengths 0..=32
    --> $FILE:1155:5
     |
1155 |     pub test: [::std::os::raw::c_char; 104usize],
     |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `std::array::LengthAtMost32` is not implemented for `[i8; 104]`
     |
     = note: required because of the requirements on the impl of `std::fmt::Debug` for `[i8; 104]`
     = note: required because of the requirements on the impl of `std::fmt::Debug` for `&[i8; 104]`
     = note: required for the cast to the object type `dyn std::fmt::Debug`

[emilio]

@martinthomson / @mmacedoeu do you have a test-case for that?

[mmacedoeu]

@martinthomson / @mmacedoeu do you have a test-case for that?

Not really a short test-case, It is a closed source code, but It seams any attempt to generate bindgen where a window.h header is present lead to this. I will have some time soon to try and isolate the problem. Is there any tool that could help automate the test-case ?

[emilio]

Yeah, you can use creduce as explained here to minimize test-cases.

[martinthomson]

The test case is from neqo, but shove this in a header:

struct Test {
  char test[104];
};

[emilio]

I cannot reproduce that, is there any chance you're using a combination of:

• An old version of bindgen (<0.51.1), and...
• A new version (1.39+) of rustc?

If so, that's #1627 / rust-lang/rust#64710.

[martinthomson]

Yeah, that's it. Thanks. It was 0.51.0. I'll get the latest.