Test suite work

[RalfJung]

Fix most compile-fail tests. Some are left ignored, but at least the rest is tested against regressions.

Also move some tests out of fullmir that no longer need it.

[RalfJung]

Panic handling on Windows is unhappy:

error[E0080]: constant evaluation error
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\sys\windows\compat.rs:33:22
   |
33 |         let handle = c::GetModuleHandleW(module.as_ptr());
   |                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ no mir for `std::sys::windows::c::::GetModuleHandleW`
   |
note: inside call to `std::sys::windows::compat::lookup`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\sys\windows\mutex.rs:150:21
   |
150|     let ret = match compat::lookup("kernel32", "AcquireSRWLockExclusive") {
   |                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
note: inside call to `std::sys::windows::mutex::kind`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\sys\windows\mutex.rs:68:15
   |
68 |         match kind() {
   |               ^^^^^^
note: inside call to `std::sys::windows::mutex::Mutex::lock`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\sys_common\mutex.rs:40:37
   |
40 |     pub unsafe fn raw_lock(&self) { self.0.lock() }
   |                                     ^^^^^^^^^^^^^
note: inside call to `std::sys_common::mutex::Mutex::raw_lock`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\sys_common\mutex.rs:46:9
   |
46 |         self.raw_lock();
   |         ^^^^^^^^^^^^^^^
note: inside call to `std::sys_common::mutex::Mutex::lock`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\sys_common\thread_local.rs:165:26
   |
165|             let _guard = INIT_LOCK.lock();
   |                          ^^^^^^^^^^^^^^^^
note: inside call to `std::sys_common::thread_local::StaticKey::lazy_init`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\sys_common\thread_local.rs:152:18
   |
152|             0 => self.lazy_init() as imp::Key,
   |                  ^^^^^^^^^^^^^^^^
note: inside call to `std::sys_common::thread_local::StaticKey::key`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\sys_common\thread_local.rs:140:52
   |
140|     pub unsafe fn get(&self) -> *mut u8 { imp::get(self.key()) }
   |                                                    ^^^^^^^^^^
note: inside call to `std::sys_common::thread_local::StaticKey::get`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\thread\local.rs:443:23
   |
443|             let ptr = self.os.get() as *mut Value<T>;
   |                       ^^^^^^^^^^^^^
note: inside call to `<std::thread::Key<T>><std::cell::Cell<usize>>::get`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\thread\local.rs:188:17
   |
188|                 __KEY.get()
   |                 ^^^^^^^^^^^
note: inside call to `std::rt::update_panic_count::PANIC_COUNT::__getit`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\thread\local.rs:291:24
   |
291|             let slot = (self.inner)().ok_or(AccessError {
   |                        ^^^^^^^^^^^^^^
note: inside call to `<std::thread::LocalKey<T>><std::cell::Cell<usize>>::try_with::<[closure@DefId(1/1:1801 ~ std[4c00]::panicking[0]::update_panic_count[0]::{{closure}}[0]) 0:&isize], usize>`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\thread\local.rs:248:9
   |
248|         self.try_with(f).expect("cannot access a TLS value during or \
   |         ^^^^^^^^^^^^^^^^
note: inside call to `<std::thread::LocalKey<T>><std::cell::Cell<usize>>::with::<[closure@DefId(1/1:1801 ~ std[4c00]::panicking[0]::update_panic_count[0]::{{closure}}[0]) 0:&isize], usize>`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\panicking.rs:240:5
   |
240| /     PANIC_COUNT.with(|c| {
241| |         let next = (c.get() as isize + amt) as usize;
242| |         c.set(next);
243| |         return next
244| |     })
   | |______^
note: inside call to `std::rt::update_panic_count`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\panicking.rs:449:18
   |
449|     let panics = update_panic_count(1);
   |                  ^^^^^^^^^^^^^^^^^^^^^
note: inside call to `std::panicking::rust_panic_with_hook`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\panicking.rs:390:5
   |
390| /     rust_panic_with_hook(
391| |         &mut PanicPayload::new(msg),
392| |         info.message(),
393| |         &file_line_col);
   | |_______________________^
note: inside call to `std::panicking::continue_panic_fmt`
  --> C:\Users\appveyor\.rustup\toolchains\nightly-i686-pc-windows-msvc\lib\rustlib\src\rust\src\libstd\panicking.rs:345:5
   |
345|     continue_panic_fmt(&info)
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^
note: inside call to `std::rt::begin_panic_fmt`
  --> tests/compile-fail\panic.rs:6:5
   |
6  |     assert_eq!(5, 6);
   |     ^^^^^^^^^^^^^^^^^
   = note: this error originates in a macro outside of the current crate (in Nightly builds, run with -Z external-macro-backtrace for more info)

[oli-obk]

We can merge without windows working. Or just ignore the tests on windows for now. Once my const panic PR is merged we can copy the logic and do something similar and OS independent

[RalfJung]

Once my const panic PR is merged we can copy the logic and do something similar and OS independent

Ah, good point. I was going to try and add some more shims for these syscalls (or whatever they are), but if this will all change soon anyway, I'll just make it ignore the test.

I'd prefer to keep AppVeyor green.