Windows fixes

[bjorn3]

Fixes the issues found in https://github.com/bjorn3/rustc_codegen_cranelift/issues/1193. This still needs testing on CI.
cc @EndilWayfare

[EndilWayfare]

./test.sh initially fails with

[BUILD] mini_core
C:/Users/{redacted}/rustc_codegen_cranelift/build/bin/cg_clif.exe: error while loading shared libraries: rustc_codegen_cranelift.dll: cannot open shared object file: No such file or directory

Copying rustc_codegen_cranelift.dll to the repo root fixes that. :/

---

Then it fails with

[BUILD] mini_core
C:/Users/{redacted}/rustc_codegen_cranelift/build/bin/cg_clif.exe: error while loading shared libraries: std-77774f43f6f43ae2.dll: cannot open shared object file: No such file or directory

Weird, the toolchain's not on the path? Adding

export RUST_SYSROOT="$(readlink -f ~)/.rustup/toolchains/nightly-2021-08-25-x86_64-pc-windows-gnu"
PATH="$RUST_SYSROOT/bin:$PATH"

at the begging of test.sh fixes that. Strangely, it also fixes the rustc_codegen_cranelift.dll problem above; it's not necessary to copy the .dll, or even add build/bin to PATH, apparently...

---

Then it fails with

[BUILD] rustc_codegen_cranelift
// ...
   Compiling rustc_codegen_cranelift v0.1.0 (C:\Users\{redacted}\rustc_codegen_cranelift)
    Finished release [optimized] target(s) in 1.12s
error: environment variable `RUSTUP_TOOLCHAIN` not defined
  --> scripts/cargo-clif.rs:37:38
   |
37 |     env::set_var("RUSTUP_TOOLCHAIN", env!("RUSTUP_TOOLCHAIN"));
   |                                      ^^^^^^^^^^^^^^^^^^^^^^^^
   |
   = note: this error originates in the macro `env` (in Nightly builds, run with -Z macro-backtrace for more info)

export RUSTUP_TOOLCHAIN=nightly-2021-08-25-x86_64-pc-windows-gnu

---

Then `[TEST] rust-random/rand` fails:

     Running unittests (target\debug\deps\rand_distr-97e69b0a61466c32.exe)

running 50 tests
test dirichlet::test::test_dirichlet_invalid_length - should panic ... ok
test binomial::test::test_binomial_invalid_lambda_neg - should panic ... ok
// ...
test pareto::tests::sample ... ok
test pareto::tests::value_stability ... FAILED
test pert::test::test_pert ... ok
// ...
test weighted_alias::test::test_weighted_index_u8 ... ok
test weighted_alias::test::test_weighted_index_i128 ... ok
test weighted_alias::test::test_weighted_index_u128 ... ok

failures:

---- pareto::tests::value_stability stdout ----
---- pareto::tests::value_stability stderr ----
thread 'main' panicked at 'assertion failed: `(left == right)`
  left: `[9.019295276219138, 4.3097126018270595, 6.837815045397156, 105.8826669383772]`,
 right: `[9.019295276219136, 4.3097126018270595, 6.837815045397157, 105.8826669383772]`', rand_distr\src\pareto.rs:119:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    pareto::tests::value_stability

test result: FAILED. 49 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.23s

Huh. Well that's certainly odd. I have no idea why rand isn't producing stable values, so that's where I'm stuck.

[bjorn3]

Closing as outdated. Some parts have been fixed through the rewrite of the test harness in rust (by @afonso360) the rest had already landed on master I believe. There are still a couple of remaining issues for full Windows support though.