[SR-13820] Crasher in StdlibUnittest.RangeSet #56218
Description
| Previous ID | SR-13820 |
| Radar | None |
| Original Reporter | michellecasbon (JIRA User) |
| Type | Bug |
| Status | Resolved |
| Resolution | Done |
Environment
Ubuntu 18.04
Additional Detail from JIRA
| Votes | 0 |
| Component/s | Compiler |
| Labels | Bug |
| Assignee | @meg-gupta |
| Priority | Medium |
md5: c0e8d9509c991624bc8fd7d78a9dcc3d
blocks:
Issue Description:
Bisection reveals that #34188 causes a core dump in the StdlibUnittest.RangeSet test in S4TF, which looks to be a miscompile.
To reproduce in Ubuntu 18.04, use the tensorflow branch at 1e86c04.
1. Checkout compatible versions of auxiliary repos:
swift/utils/update-checkout --skip-repository swift --clone --scheme tensorflow2. Run a build:
swift/utils/build-script --preset tensorflow_testAll tests pass at the commit that adds #30710.
3. Switch to one commit prior, delete the build/Ninja-ReleaseAssert/swift-linux-x86_64 directory, and build again.
git -C swift checkout f8a8d087bc498448b135c52df4d6a3c8485a9fca
rm -rf build/Ninja-ReleaseAssert/swift-linux-x86_64
swift/utils/build-script --preset tensorflow_testThe StdlibUnittest.RangeSet test now fails.
The test does not fail individually, only when run as part of a suite:
cd build/Ninja-ReleaseAssert/swift-linux-x86_64
ninja check-swift-validation-optimize-linux-x86_64Debugging reveals that the problem exists in the every() function. If you replace the contents with this functional equivalent, the test passes:
func every(_ n: Int) -> [Element] {
var a: Array<Element> = []
sequence(first: startIndex) { i in
self.index(i, offsetBy: n, limitedBy: self.endIndex)
}.forEach { a.append(self[$0]) }
return a
}