Fix up tests for Windows #5074
Fix up tests for Windows #5074
Conversation
|
swiftlang/swift#75932 @swift-ci please test Windows platform |
1 similar comment
|
swiftlang/swift#75932 @swift-ci please test Windows platform |
|
swiftlang/swift#75932 @swift-ci please test Windows platform |
1 similar comment
|
swiftlang/swift#75932 @swift-ci please test Windows platform |
|
swiftlang/swift#75932 @swift-ci please test Windows platform |
2 similar comments
|
swiftlang/swift#75932 @swift-ci please test Windows platform |
|
swiftlang/swift#75932 @swift-ci please test Windows platform |
|
swiftlang/swift#75932 @swift-ci please test Linux platform |
| @@ -21,7 +21,9 @@ let validPathSeps: [Character] = ["/"] | |||
| #endif | |||
|
|
|||
| public func NSTemporaryDirectory() -> String { | |||
| FileManager.default.temporaryDirectory.path() | |||
| FileManager.default.temporaryDirectory.withUnsafeFileSystemRepresentation { | |||
There was a problem hiding this comment.
This was necessary to ensure that paths with drive letters do not have a leading slash (the file system representation standardizes that)
| @@ -78,6 +78,14 @@ class BundlePlayground { | |||
| #endif | |||
| } | |||
| } | |||
|
|
|||
| var fileNameSuffix: String? { | |||
There was a problem hiding this comment.
I don't believe Foundation's tests have been run in debug mode before, so this test was failing when Foundation was built in debug mode and this fixes it by adding the appropriate file suffix
There was a problem hiding this comment.
Correct! We always build in release mode currently due to other limitations that I would like to see lifted.
| @@ -104,6 +104,8 @@ class TestThread : XCTestCase { | |||
| func test_callStackSymbols() throws { | |||
| #if os(Android) || os(OpenBSD) | |||
| throw XCTSkip("Android/OpenBSD doesn't support backtraces at the moment.") | |||
| #elseif os(Windows) | |||
| throw XCTSkip("This test is unexpectedly crashing in CI at the moment.") | |||
There was a problem hiding this comment.
I'm not certain why, but these tests were crashing in CI. Disabling for now until we can followup with a fix
| @@ -652,6 +652,9 @@ final class TestURLSession: LoopbackServerTest, @unchecked Sendable { | |||
| } | |||
|
|
|||
| func test_repeatedRequestsStress() async throws { | |||
| #if os(Windows) | |||
| throw XCTSkip("This test is currently disabled on Windows") | |||
There was a problem hiding this comment.
This was a new test added after the re-core, so the test hadn't been run on Windows CI before. It's failing and @jrflat is looking into it
|
|
||
| XCTAssertThrowsError(try fm.contentsOfDirectory(atPath: "")) { | ||
| let code = ($0 as? CocoaError)?.code | ||
| XCTAssertEqual(code, .fileReadNoSuchFile) | ||
| XCTAssertEqual(code, emptyFileNameError ?? .fileReadNoSuchFile) |
There was a problem hiding this comment.
I'm open to suggestions on this. Previously in SCL-F, for empty paths we always threw .fileReadInvalidFileName. As part of the re-core, the Linux implementation changed to match the Darwin implementation which threw errors like .fileReadNoSuchFile. However, the Windows implementation still throws .fileReadInvalidFileName. For now, this just updates the test to account for the behavior today, but if we want (potentially in a separate change if that's cleaner) we can update the Windows implementation to throw the same error codes
There was a problem hiding this comment.
Yes, please! We should match the Darwin error codes if possible. The Windows path did reference the Linux path heavily for error conditions and semantics originally and the two diverged :(.
There was a problem hiding this comment.
Filed swiftlang/swift-foundation#882 to track standardizing this as a followup (I think it'd be great to land these tests now to prevent any other behavioral regressions on Windows and I'll followup shortly with a change to swift-foundation to throw the right error code and copy these tests to that repo as well
| @@ -78,6 +78,14 @@ class BundlePlayground { | |||
| #endif | |||
| } | |||
| } | |||
|
|
|||
| var fileNameSuffix: String? { | |||
There was a problem hiding this comment.
Correct! We always build in release mode currently due to other limitations that I would like to see lifted.
This addresses some test failures where the tests need to be updated for the re-core on Windows. In some cases, the underlying code was changed and the test should be updated to reflect the new and/or correct behavior - in other cases the tests may have been lucky before but are incorrect and need to be changed or removed.