Unable to build rust torch bindings https://github.com/posix4e/rust-torch

[posix4e]

Input C/C++ Header

#include "TH/TH.h"

We actually generate it.

bindgen --no-derive-debug --builtins tmp/expanded.h -o "$output"

Actual Results

Error { repr: Custom(Custom { kind: Other, error: StringError("Cannot find binary path") }) }

http://termbin.com/lgfd

http://termbin.com/d39u

Expected Results

I expected it to compile

[posix4e]

I tried it with an older version of bindgen and I didn't get an error when compiling but i still get an error when building

posi@posi:~/src/rust-torch$ cargo build
   Compiling torch v0.1.0 (file:///home/posi/src/rust-torch)
error: const fn is unstable (see issue #24111)
   --> src/lib.rs:420:5
    |
420 | /     pub const fn new_bitfield_1(__w_termsig: ::std::os::raw::c_uint,
421 | |                                 __w_coredump: ::std::os::raw::c_uint,
422 | |                                 __w_retcode: ::std::os::raw::c_uint) -> u32 {
423 | |         let bitfield_unit_val =
...   |
441 | |         bitfield_unit_val | __w_retcode
442 | |     }
    | |_____^

error: const fn is unstable (see issue #24111)
   --> src/lib.rs:498:5
    |
498 | /     pub const fn new_bitfield_1(__w_stopval: ::std::os::raw::c_uint,
499 | |                                 __w_stopsig: ::std::os::raw::c_uint) -> u32 {
500 | |         let bitfield_unit_val =
501 | |             {
...   |
511 | |         bitfield_unit_val | __w_stopsig
512 | |     }
    | |_____^

error: aborting due to previous error(s)

error: Could not compile `torch`.

[posix4e]

Apparently with this old version of bindgen it works if i tell it not to generate unstable rust code. Should I leave this open?

[fitzgen]

Hi @posix4e, thanks for filing a bug. Can you provide a standalone version of the input header file? https://github.com/rust-lang-nursery/rust-bindgen/blob/master/CONTRIBUTING.md#using-creduce-to-minimize-test-cases

Without that, unfortunately, this this issue is not very actionable. Thanks!

[posix4e]

@fitzgen the one line h file generated http://termbin.com/lgfd

[fitzgen]

@fitzgen the one line h file generated http://termbin.com/lgfd

The "one line h file" is an include of another header, which includes an unknown further number of headers, each of which include ... etc.

In order for this issue to be actionable, we need at minimum the header after the preprocessor runs and preferrably after running creduce as well. You can find instructions for producing these things in the link I gave above.

Thanks!

[XVilka]

@fitzgen I have the same error and ran a creduce on it, sadly it returns exactly the same file after a whole night of trying to reduce it. The file is https://github.com/radare/radare2/blob/master/libr/include/r_core.h This is my creduce predicate script:

CMD="bindgen --verbose /home/user/data/tmp/r_core.h -o /home/user/data/tmp/r2-api/r_core.rs -- -I/usr/local/include/libr"
OUT=$($CMD 3>&1 1>&2 2>&3)
EXPECT="Error { repr: Custom(Custom { kind: Other, error: StringError(\"Cannot find binary path\") }) }"

if [ "$OUT" == "$EXPECT" ]; then
	exit 0
else
	exit 1
fi

Here is the preprocessed file
__bindgen.i.gz

[XVilka]

@fitzgen should I open a separate bug for this radare2 case? Or it's fine here? Also creduce failed on this testcase, I sent them bugreport, they will check it.