skeptic + cross + *-pc-windows-gnu = Bad EXE format

[slonopotamus]

Problematic build is here: https://travis-ci.org/slonopotamus/local-encoding-rs/jobs/427679330#L673

It is run via cargo install cross && cross test --target i686-pc-windows-gnu. Note that everything works natively on both Linux and Windows.

I am actually not sure whether this is skeptic or cross problem (or mine).

Commit f5b4e2e gives the impression that skeptic is supposed to work with cross.

[slonopotamus]

I think understood what's going on. Skeptic tries to run cargo via wine. But cargo is a binary for host platform, Linux x86_64.

[slonopotamus]

So. I think this is a fundamental flaw in skeptic design - skeptic attempts to compile test body during tests run. But, you cannot expect that cargo, rustc or other compiling stuff is available when tests are run.

This is what happens:

1. Skeptic-generated test is run
2. It calls skeptic::rt::run_test
3. Which in turn calls skeptic::rt::compile_test_case (this is where problems start)
4. Which in turn calls skeptic::rt::get_rlib_dependencies
5. Which in turn calls skeptic::rt::LockedDeps::from_path
6. Which in turn calls cargo_metadata::metadata_deps
7. Which in turn calls cargo_metadata::metadata_run
8. Which attempts to run cargo (via wine in my case because tests are run within wine, this is a target platform)

But there is no cargo on target platform. Test is supposed to be a standalone binary that can just be run on whatever platform library targets for. That platform may or may not be able to compile binaries itself.

[slonopotamus]

#78 fixes my problem, but #78 has a discussion with reasons why it isn't going to be accepted.