[Test] Disable StaticBigInt test on watchOS

[xymus]

No description provided.

[tbkka]

This seems to be an issue because the test verifies some things on Int and UInt, but doesn't account for the fact that those types are 32 bits on some platforms.

CC: @benrimmington

[benrimmington]

@tbkka Thanks for the ping, I was unaware that this test file had been disabled.

(Although the // UNSUPPORTED: comment might be ignored, if it appears after the // END. comment.)

I'm unable to test on a watchOS device or 32-bit simulator. Please could you share the output of the test failures?

[tbkka]

Here you go:

.../swift/test/stdlib/StaticBigInt.swift:181:21: error: integer literal '4822678189205111' overflows when stored into 'UInt'
        expectEqual(0x00112233_44556677, positive.actual[1])
                    ^
.../swift/test/stdlib/StaticBigInt.swift:180:21: error: integer literal '9843086184167632639' overflows when stored into 'UInt'
        expectEqual(0x8899AABB_CCDDEEFF, positive.actual[0])
                    ^
.../swift/test/stdlib/StaticBigInt.swift:165:21: error: integer literal '18446744073709551615' overflows when stored into 'UInt'
        expectEqual(0xFFFFFFFF_FFFFFFFF, negative.actual[.max])
                    ^
.../swift/test/stdlib/StaticBigInt.swift:164:21: error: integer literal '18446744073709551615' overflows when stored into 'UInt'
        expectEqual(0xFFFFFFFF_FFFFFFFF, negative.actual[2])
                    ^
.../swift/test/stdlib/StaticBigInt.swift:163:21: error: integer literal '18441921395520346504' overflows when stored into 'UInt'
        expectEqual(0xFFEEDDCC_BBAA9988, negative.actual[1])
                    ^
.../swift/test/stdlib/StaticBigInt.swift:162:21: error: integer literal '8603657889541918977' overflows when stored into 'UInt'
        expectEqual(0x77665544_33221101, negative.actual[0])
                    ^

[benrimmington]

@tbkka Thanks. Those statements are within else if getInt(UInt.bitWidth) == 64 branches, so they shouldn't be executing on a watchOS device.

[tbkka]

Were those conditionals added at some point? Maybe this test just got unlucky and ran before that?

[benrimmington]

Were those conditionals added at some point?

No, the tests haven't changed since they were first merged two weeks ago.

The only recent change was to add a feature flag in #62541, but that didn't touch the tests.

[tbkka]

That's ... confusing. I'll try to dig out some more details and see if I can narrow down the problem here.

One question: Why getInt()?? It's not obvious to me why the conditions here aren't just Uint.bitWidth == 64.

[benrimmington]

One question: Why getInt()??

To avoid "warning: will never be executed"; otherwise the compiler can see the @_transparent implementation.

https://github.com/apple/swift/blob/77527c9e41b3d71501bf35fe85bf0a4d58b7fcce/stdlib/public/core/IntegerTypes.swift.gyb#L1355-L1365

[tbkka]

Oh. I see. Those tests aren't executing on watchOS. This is a compile-time failure.

[benrimmington]

This is a compile-time failure.

OK, so the tests should be using:

- if getInt(UInt.bitWidth) == 32 {
+ #if arch(i386) || arch(arm) || arch(arm64_32) || arch(wasm32) || arch(powerpc)

- } else if getInt(UInt.bitWidth) == 64 {
+ #elseif arch(x86_64) || arch(arm64) || arch(powerpc64) || arch(powerpc64le) || arch(s390x)

- if getInt(UInt.bitWidth) == 64 {
+ #if arch(x86_64) || arch(arm64) || arch(powerpc64) || arch(powerpc64le) || arch(s390x)

[benrimmington]

The main and release/5.8 branches will be locked tomorrow at 10 AM PST for the holidays. Can this be fixed and approved in time?

[tbkka]

If you can get the PR up soon enough, we can certainly try.