[SR-12204] "Remangling failed" on Linux 5.1 and 5.2

[tachyonics]

Previous ID	SR-12204
Radar	rdar://problem/59496022
Original Reporter	@tachyonics
Type	Bug
Status	Resolved
Resolution	Done

Attachment: Download

Environment

The following docker images can reproduce this bug- swift:5.1.4-bionic: bug reproduced swift:5.1.4-xenial: bug reproduced swift:5.0.3-bionic: bug doesn't occur swift:5.0.3-xenial: bug doesn't occur swiftlang/swift:nightly-master-xenial: bug reproduced swiftlang/swift:nightly-5.2-xenial: bug reproduced

Additional Detail from JIRA

| | | |------------------|-----------------| |Votes | 0 | |Component/s | | |Labels | Bug | |Assignee | None | |Priority | Medium | md5: 144d64d07e4f10ebf45d3458ebd0933a

Issue Description:

This bug can be replicated using a development branch of the SmokeFramework package-

https://github.com/amzn/smoke-framework/tree/request_tracing

Specially when this line is added-

https://github.com/amzn/smoke-framework/blob/request_tracing/Sources/SmokeOperationsHTTP1/SmokeHTTP1Server%2BstartAsOperationServer.swift#L42-L47

An example of the compile failure is here-

https://travis-ci.com/amzn/smoke-framework/jobs/287439552#L493

[beccadax]

@swift-ci create

[eeckstein]

@tachyonics I could not reproduce the issue.
Which toolchain version did you use?

[tachyonics]

I was able to reproduce the issue using the 5.1.4-bionic and 5.1.4-xenial docker images plus nightly-5.2-xenial and nightly-master-xenial docker images.

5.1.4-bionic: https://travis-ci.com/amzn/smoke-framework/jobs/287439552

5.1.4-xenial: https://travis-ci.com/amzn/smoke-framework/jobs/287439553

nightly-5.2-xenial: https://travis-ci.com/amzn/smoke-framework/jobs/287439558

nightly-master-xenial: https://travis-ci.com/amzn/smoke-framework/jobs/287439556

I didn't encounter the issue when compiling on the mac.

[eeckstein]

Interesting.
I tried with a top-of-master compiler and could not reproduce it.
How did you build? Debug or release configuration?

[tachyonics]

Looks like it is only reproducible in Debug configuration.

[gwynne]

I just hit this with the 5.2-RELEASE compiler. This snippet is sufficient to reproduce the crash by itself:

extension Optional: Comparable where Wrapped: Comparable {
    public static func v(in r: ClosedRange<Self>) {}
    public static func < (lhs: Self, rhs: Self) -> Bool { false }
}

Remangling failed:
original     = $sSNyxSgxSgSL13remangle_failxSLHD1__HCg_GD
remangled    = $sSNyxSgAASL13remangle_failxSLHD1__HCg_GD
Stack dump:
0.  Program arguments: /usr/bin/swift -frontend -c -primary-file /remangle-fail/Sources/remangle-fail/main.swift -emit-module-path /remangle-fail/.build/x86_64-unknown-linux-gnu/debug/remangle_fail.build/main~partial.swiftmodule -emit-module-doc-path /remangle-fail/.build/x86_64-unknown-linux-gnu/debug/remangle_fail.build/main~partial.swiftdoc -emit-module-source-info-path /remangle-fail/.build/x86_64-unknown-linux-gnu/debug/remangle_fail.build/main~partial.swiftsourceinfo -emit-dependencies-path /remangle-fail/.build/x86_64-unknown-linux-gnu/debug/remangle_fail.build/main.d -emit-reference-dependencies-path /remangle-fail/.build/x86_64-unknown-linux-gnu/debug/remangle_fail.build/main.swiftdeps -target x86_64-unknown-linux-gnu -disable-objc-interop -I /remangle-fail/.build/x86_64-unknown-linux-gnu/debug -color-diagnostics -enable-testing -g -module-cache-path /remangle-fail/.build/x86_64-unknown-linux-gnu/debug/ModuleCache -swift-version 5 -Onone -D SWIFT_PACKAGE -D DEBUG -enable-anonymous-context-mangled-names -module-name remangle_fail -o /remangle-fail/.build/x86_64-unknown-linux-gnu/debug/remangle_fail.build/main.swift.o -index-store-path /remangle-fail/.build/x86_64-unknown-linux-gnu/debug/index/store -index-system-modules
1.  Swift version 5.2 (swift-5.2-RELEASE)
2.  While emitting IR SIL function "@$sSq13remangle_failSLRzlE1v2inySNyxSgADSLAAxSLHD1__HCg_G_tFZ".
 for 'v(in:)' (at /remangle-fail/Sources/remangle-fail/main.swift:2:12)
3.  While mangling type for debugger type 'ClosedRange<Optional<Wrapped>>'

Confirmed it only shows up in Debug builds on Linux. Will also attach repro as test project in package form momentarily.

[slavapestov]

The bug might only show up on Linux if there's something about debug info that gets emitted differently on the two platforms. The mangled name here is coming from debug info emission.

It appears that in the original ASTMangler output, we have two occurrences of xSg. In the remangled version, the second one becomes a substitution AA. So I think the bug is that ASTMangler does not record a substitution in some case, but the remangler does. Unless this only comes up with the DWARF mangling, we have to say that the ASTMangler is "correct", since changing it will break the ABI. So we should change the remangler to not record a substitution here. It's important that the ASTMangler and remangler produce identical output.

[slavapestov]

https://github.com/apple/swift/pull/30840