[StdLib] Extend ReflectionHashing test to pass on Linux

[ezephir]

Add a test case for X86_64 platforms that do not use Objective-C interop
and removes the XFAIL on Linux. Partially resolves SR-216.

This change passes the normal test suite on OS X 10.10 and Unbuntu 14.04.

The core reason the tests fail is because String is hashed differently depending on whether or not the platform supports the Obj-C runtime. Foundation's hashcode is used when the runtime is present and the built-in hashcode function is used otherwise. This seems reasonable, so I've extended the test with a new case.

[moiseev]

Thanks! 👍

[dcci]

Sorry to rain on your parade, but this broke my Fedora buildbot. Can you please take a look?

FAIL: Swift :: 1_stdlib/ReflectionHashing.swift (88 of 2415)
******************** TEST 'Swift :: 1_stdlib/ReflectionHashing.swift' FAILED ********************
Script:
--
/home/davide/swift/build/Ninja-ReleaseAssert/swift-linux-x86_64/bin/swiftc -target x86_64-unknown-linux-gnu  -module-cache-path '/tmp/swift-testsuite-clang-module-cache5Jq4Zn'   -module-name a /home/davide/swift/swift/test/1_stdlib/ReflectionHashing.swift -o /home/davide/swift/build/Ninja-ReleaseAssert/swift-linux-x86_64/test-linux-x86_64/1_stdlib/Output/ReflectionHashing.swift.tmp.out
/home/davide/swift/build/Ninja-ReleaseAssert/swift-linux-x86_64/test-linux-x86_64/1_stdlib/Output/ReflectionHashing.swift.tmp.out
--
Exit Code: 134

Command Output (stdout):
--
[ RUN      ] Reflection.Dictionary/Empty
[       OK ] Reflection.Dictionary/Empty
[ RUN      ] Reflection.Dictionary
out>>> check failed at /home/davide/swift/swift/test/1_stdlib/ReflectionHashing.swift, line 97
out>>> expected: "▿ 5 key/value pairs\n  ▿ [0]: (2 elements)\n    - .0: One\n    - .1: 1\n  ▿ [1]: (2 elements)\n    - .0: Five\n    - .1: 5\n  ▿ [2]: (2 elements)\n    - .0: Two\n    - .1: 2\n  ▿ [3]: (2 elements)\n    - .0: Four\n    - .1: 4\n  ▿ [4]: (2 elements)\n    - .0: Three\n    - .1: 3\n" (of type Swift.String)
out>>> actual: "▿ 5 key/value pairs\n  ▿ [0]: (2 elements)\n    - .0: One\n    - .1: 1\n  ▿ [1]: (2 elements)\n    - .0: Three\n    - .1: 3\n  ▿ [2]: (2 elements)\n    - .0: Four\n    - .1: 4\n  ▿ [3]: (2 elements)\n    - .0: Five\n    - .1: 5\n  ▿ [4]: (2 elements)\n    - .0: Two\n    - .1: 2\n" (of type Swift.String)
[     FAIL ] Reflection.Dictionary
[ RUN      ] Reflection.Set
[       OK ] Reflection.Set
Reflection: Some tests failed, aborting
UXPASS: []
FAIL: ["Dictionary"]
SKIP: []
To debug, run:
$ /home/davide/swift/build/Ninja-ReleaseAssert/swift-linux-x86_64/test-linux-x86_64/1_stdlib/Output/ReflectionHashing.swift.tmp.out --stdlib-unittest-in-process --stdlib-unittest-filter "Dictionary"

[dcci]

I'm not familiar with the internal of stdlib but the obvious culprit seems to be that the result of hashing is different.

[ezephir]

@dcci That's exactly the culprit for why the original test failed before I extended it. I'll look into it.

[ezephir]

Status update: Still looking at this. So far, no luck.

So far I've confirmed that it still works on my particular install of Ubuntu 14.04 with a full, non-incremental build. I've also explored language and locale settings, but I believe that the test runner filters these out and that stdlib ignores them for the purposes of hashing.

[ezephir]

Status update: I've recently been able to reproduce the under Fedora 23. I'm still debugging the issue to see why the hashes are different.

[ezephir]

@dcci I've reverted this change in PR #933 and have started a dialogue there on fixing it properly. Thank you for noticing this break and for your patience.