* vm.rs - calling Ctx data_finalizer upon destruction

[YaronWittenstein]

It seems that the data_finalizer field of the Ctx is never being called upon instance destruction.

It seems a bit tricky to assert that in though

@syrusakbary @Hywan

[syrusakbary]

Ummm... tests are broken on windows, so we have to take a deeper look.
Will trigger bors to see the results on macos/linux.

[syrusakbary]

bors try

[bors]

try

Build failed

• ci/circleci: test-stable

[syrusakbary]

(still investigating)

[syrusakbary]

bors try

[bors]

try

Build failed

• ci/circleci: test-stable

[MarkMcCaskey]

Hey @YaronWittenstein I found the issue on our side! We are breaking the contract on std::mem::uninitialized() and calling Drop on garbage memory interpreted as a Ctx.

The reason is that we do something like

let ctx = Box::new(std::mem::uninit());
something_else.val = Box::into_raw(ctx);
something_else.val = real_ctx; // this is invalid because it causes the uninit Ctx to Drop

And the solution is to change the last line to

let mut real_ctx = real_ctx;
std::mem::swap(&mut real_ctx, &mut something_else.val);
std::mem::forget(real_ctx); // tell Rust to not Drop our garbage Ctx

I have a fix locally that I need to clean up -- and actually I need to port it to the new MaybeUninit style for Rust 1.38 -- but your PR was passing all the tests I tried on my computer last night.

Thanks! I'll ping you again when its merged into master

[MarkMcCaskey]

bors try

[bors]

try

Build succeeded

• ci/circleci: test-macos
• continuous-integration/appveyor/branch
• ci/circleci: lint
• ci/circleci: test-stable
• ci/circleci: test