Missing lifetime specifier in generated code

[SKyletoft]

Expected Behavior

There would be a lifetime specifier in the generated function

Actual Behavior

Borrowchecking error
Error log

Steps to Reproduce the Problem

1. Include this header
2. generate! the type s2e::S2E

Reproduced in this repo

Specifications

• Version: autocxx 0.24.0, Rust 1.66.1
• Platform: Linux NixOS Unstable x64 (nixpkgs), Ubuntu 22.04 x64 (rustup)

[adetaylor]

Thanks, I'm traveling this week but I'll take a look next week. It would be even better if you could raise a PR with a failing integration test.

[SKyletoft]

I've been trying to figure it out. I have this command (running in an Ubuntu VM with the repo linked above):

autocxx-reduce \
	--problem="error[E0106]" \
	--clang-arg="-std=gnu++17" \
	--clang-arg="-DBOOST_BIND_GLOBAL_PLACEHOLDERS=1" \
	--clang-arg="-DTARGET_PAGE_BITS=12" \
	--clang-arg="-DSE_RAM_OBJECT_BITS=12" \
	-k \
	file \
	-d "safety!(unsafe_ffi)" \
	-d 'generate!("s2e::S2E")' \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/klee/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libcoroutine/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libcpu/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libfsigc++/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libq/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libs2e/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libs2ecore/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libs2eplugins/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libtcg/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/libvmi/include/" \
	-I "/home/u3836/git/s2e-autocxx-error/s2e/tools/lib/X8664BitcodeLibrary" \
	--header "/home/u3836/git/s2e-autocxx-error/s2e/libs2ecore/include/s2e/S2E.h"

I had to patch autocxx_engine to use emit a -std=gnu++17 because passing that flag via --clang-arg doesn't seem to be respected by clang given that it follows a -std=c++14.

But this still spits out slightly over 200k lines of C++, so I assume I'm missing something.

[adetaylor]

Thanks, yeah, autocxx-reduce (when it's working) will shrink that down to just 2-3 lines of code! However, it's quite pernickety to get working, because it has to make assumptions about your local build environment (as you noticed).

Let me see if I can reproduce it using the example repo you gave.

[adetaylor]

I can reproduce this with your provided standalone repo (thanks for doing that), I made a repro.json and I'm now running it through the reduction pipeline. From here on it's just a matter of time (possibly a week) until it spits out a minimized test case.

[adetaylor]

Here is the minimized test case spat out by the reduction pipeline:

namespace a {
class b;
}
namespace s2e {
class c;
class f {
  a::b d();
};
class S2E {
public:
  f e;
  a::b &d(c *) const;
};
} // namespace s2e

I expect it can be trimmed down a little more by hand, but that should give me enough to work with to diagnose the problem. I'm in the middle of a few other things right now but will take a look eventually. Thanks for the report, and thanks for making it possible to get a minimal test case like this.

[adetaylor]

I think I see what's going on here.

The generated code (about which rustc complains) is:

pub unsafe fn S2E_d_autocxx_wrapper_0xa21e7b0ce9f73499 (autocxx_gen_this : & S2E , arg1 : * mut c) -> Pin < & mut b > ;

The code generator assumed that lifetimes could be elided in this case, but they can't, because the the input reference is & and the output reference is &mut. Of course, explicit lifetimes also can't help here - I think we need to reject such functions, at least until all the CppRef stuff enables us to move beyond normal Rust reference semantics.