cross compiling using bindgen under x86_64, but requires 32 bit

[sprhawk]

I'm building btrfsutil-rs,
it complains following errors:

| --- stdout
| cargo:rustc-link-lib=btrfsutil
| cargo:rerun-if-changed=btrfsutil_wrapper.h
| cargo:rerun-if-changed=bindgen_whitelist.toml
|
| --- stderr
| /usr/include/x86_64-linux-gnu/gnu/stubs.h:7:11: fatal error: 'gnu/stubs-32.h' file not found
| /usr/include/x86_64-linux-gnu/gnu/stubs.h:7:11: fatal error: 'gnu/stubs-32.h' file not found, err: true
| thread 'main' panicked at 'Unable to generate bindings: ()', deps/btrfsutil-rs/build.rs:44:39
| stack backtrace:

I guess is compiler when running build.rs is looking for 32bit lib, but why?

I can build ffi for libimobiledevice without problem

[sprhawk]

I have used BINDGEN_EXTRA_CLANG_ARGS to fixed it.

Is this a bug or intention ?

[nastevens]

Can you share a minimal example that demonstrates this please?

[sprhawk]

I was busy doing project. I'll try to make a small example project recently

[hellow554]

Same issue here 🙋

In my case, I have a program, that depends on socketcan-alt, which uses bindgen, to produce bindings for the <linux/can.h> header.

Here's the error message for me (which is somewhat similar to @sprhawk's)

| error: failed to run custom build command for `socketcan-alt v0.2.2 (https://github.com/Hakuyume/socketcan-rs.git?rev=f79cef958e7f4ec68ee1e0f30a7b66df15411ad6#f79cef95)`
| 
| Caused by:
|   process didn't exit successfully: `/home/op/y_unic/build/tmp/work/cortexa7t2hf-neon-vfpv4-mut-linux-gnueabi/emv-test/1.0-r0/target/release/build/socketcan-alt-31d2e4f812a3f006/build-script-build` (exit code: 101)
|   --- stderr
|   /usr/include/linux/types.h:5:10: fatal error: 'asm/types.h' file not found
|   /usr/include/linux/types.h:5:10: fatal error: 'asm/types.h' file not found, err: true

I'm building for an arm device, so like sprhawk not for the same platform. Maybe that's the issue?

A question, which you might be able to answer: What target is bindgen aiming at? The host or target platform, when he's looking for the <linux/can.h> file (e.g. does he try to use the recipe-sysroot or recipse-sysroot-native directory?)

[hellow554]

It seems to me, that he uses my systems /usr/include/linux/types.h file, instead of the sysroot ones. Is that possible? I don't think, that that is inteded, right?

[ryankurte]

this appears to be a common issue with bindgen, may be related to the "Pre-built bindgen outputs often do not match target architecture sizes" section here.

[sprhawk]

use following defines in layer to fixed it:

export BINDGEN_EXTRA_CLANG_ARGS
BINDGEN_EXTRA_CLANG_ARGS = "--sysroot=${WORKDIR}/recipe-sysroot -I${WORKDIR}/recipe-sysroot/usr/include"

[hellow554]

Hey @sprhawk
That looks like a proper fix for this Problem. IMHO this should be included into this layer.
What do you think @nastevens ?

[nastevens]

@hellow554 I'm not sure that it belongs directly in this layer, but I am more than happy to accept a PR to add something to the README since I suspect that there is a lot of crossover between people using Yocto and needing to interface Rust with C.

If this were to be included in the layer I'd rather see it as a .bbclass specifically for supporting bindgen. I don't want to lose track of this idea so I've opened #96. Please feel free to comment on that if you have specific ideas of what would be useful!