Runtime: Fix undersized allocation for out-of-line optional cast result.

[jckarter]

We accounted for the size difference between T and T? in the inline case, but not the out-of-line case. rdar://problem/27671131

[jckarter]

@swift-ci Please ASAN test

[jckarter]

@rjmccall, Does this look good for Swift 3?

[gribozavr]

Could you add a targeted test for this issue? Because it only broke in one test, we only have incidental coverage for this codepath.

[rjmccall]

Yes, this looks fine to me. I agree that it's easier to just unconditionally allocate an extra byte than to worry about whether there are extra inhabitants.

[jckarter]

Yeah, it might be cleaner to summon the Optional<T> metadata and use its allocateBuffer witness to do this, but this seemed like a good minimal fix to get the bots clean again.

[gottesmm]

This needs tests.

[jckarter]

It's exercised by the existing test suite. The tests that were failing on the ASAN bot appear to be passing now:

PASS: Swift(macosx-x86_64) :: stdlib/Dictionary.swift (8019 of 8482)
PASS: Swift(macosx-x86_64) :: stdlib/Set.swift (8056 of 8482)

[gottesmm]

@jckarter Ok on the test... On another note, I don't think that the PR asan test works now. I think xin was looking into it but it is not clear to me if it was fixed.

[trentxintong]

I do not believe the ASAN external bots are fixed. Those are machines not controlled by us and the radar got pushed back because of WWDC and other releases.

[gribozavr]

It's exercised by the existing test suite.

These is incidental coverage. If someone changes how the infrastructure of those tests work, we will lose this coverage.

[tkremenek]

@swift-ci test

[tkremenek]

@swift-ci test linux

[tkremenek]

@swift-ci test linux

[tkremenek]

@swift-ci test os x

[tkremenek]

@swift-ci test os x

[tkremenek]

@swift-ci test os x

[tkremenek]

@swift-ci test os x

[tkremenek]

ASAN testing timed out.

OS X testing failure may be infrastructure. Re-running with wiped workspace.

[tkremenek]

From OS X:

/Users/buildnode/jenkins/workspace/swift-PR-osx/buildbot_incremental/swift-macosx-x86_64/test-iphonesimulator-i386/Serialization/Output/failed-clang-module.swift.tmp/MixModA.framework/Modules/module.modulemap:3:10: error: inferred submodules require a module with an umbrella
  module * { export * }
         ^
/Users/buildnode/jenkins/workspace/swift-PR-osx/buildbot_incremental/swift-macosx-x86_64/test-iphonesimulator-i386/Serialization/Output/failed-clang-module.swift.tmp/MixModA.framework/Modules/module.modulemap:3:10: error: inferred submodules require a module with an umbrella
  module * { export * }
         ^

[tkremenek]

@swift-ci test os x platform

[jckarter]

@tkremenek The ASAN bot at least tested the tests we were interested in. OS X is still failing on the same modulemap error, but that's almost definitely unrelated.

[jckarter]

@gribozavr Crafting a test that would reliably fail outside of asan will also require depending on incidental details of the allocator, AFAICT.

[gribozavr]

@jckarter I'm not asking for a test that would fail outside of ASan -- failing only under ASan is fine (that's why we have that bot); just for a test that would cover this codepath, and that is not an emergent behavior in other tests.