Fix borrow scope for store in WrappedCallable impl for WasmtimeFn.

[peterhuene]

This PR fixes the borrow scope of store in the WrappedCallable impl of
WasmTimeFn such that it does not remain borrowed across the call to wasmtime_call_trampoline. By limiting the scope of the borrow, the implementation can be reentered if an exported function calls an imported function, which in turn calls another exported function.

Fixes #365.

[sunfishcode]

This looks good; would it be feasible to convert your example from #365 into a testcase? wasmtime-api/examples/gcd.rs may be useful as a starting point.

[peterhuene]

I didn't see formal tests for the API crate yet. Should I add a unit test or an test case as an example?

[sunfishcode]

I think it makes sense to add a tests directory, and add this as a test there.

[peterhuene]

@sunfishcode I've amended the commit with a test case.

[peterhuene]

@sunfishcode @yurydelendik It appears that there's a calling convention mismatch for the import in the new test case on Windows. Is there something I should be doing in the Rust API or is this a current limitation?

[yurydelendik]

It appears that there's a calling convention mismatch for the import in the new test case on Windows. Is there something I should be doing in the Rust API or is this a current limitation?

Let's disable tests on Windows, so it will not block landing of this PR -- I'll look into that calling convention problem shortly.

[peterhuene]

Should I file a new issue to track the calling convention mismatch panic?

[peterhuene]

Filed as #378.

[peterhuene]

cache::tests::test_write_read_cache from wasmtime-environ is failing on Windows and appears unrelated to these changes. Are there any known issues with this test?

[sunfishcode]

@mrowqa Would you mind taking a look at this test failure?

[peterhuene]

@mrowqa the cache::tests::test_write_read_cache failure is from this build: https://dev.azure.com/CraneStation/Wasmtime/_build/results?buildId=1064.

[mrowqa]

@peterhuene I can't reproduce the CI failure. I have updated the Rust version in the meantime, though. Since the test changed in the meantime, my suggestion was to rebase your PR onto master, but I can see you already did that when I was writing this comment 😃 . Other weird thing is the stack trace didn't make sense. It pointed to an empty line.

[peterhuene]

@mrowqa thanks for looking into it. I'll keep an eye out in case it fails again.

[mrowqa]

Looks like everything works. The formatting problem is within lightbeam. I have already opened an PR there CraneStation/lightbeam#28.

[sunfishcode]

Thanks!