Were you able to bisect this to a specific LLDB commit?

Comment by Sarvesh Tamba (JIRA)

@adrian-prantl , I tried to go through the recent commits for 5.1 branch, but could not make much if any of the changes could affect these 3 tests. Do you suspect any LLDB commit?

Comment by Sarvesh Tamba (JIRA)

@adrian-prantl any inputs on this one? I am not able to figure what exactly is the issue, any help would greatly be appreciated.

Comment by Sarvesh Tamba (JIRA)

I have set up the folliwing jobs at "https://ci-external.swift.org/" for building Swift 5.1 branch toolchain on PPC64LE.

• Swift - ppc64le (Tools RA, Stdlib RD, No test suites) (swift-5.1-branch)

• Swift - ppc64le (Tools RA, Stdlib RD, All test suites) (swift-5.1-branch)

The job for just building Swift 5.1 toolchain without any test suite execution completes successfully:-
https://ci-external.swift.org/job/oss-swift-5.1-RA-linux-ubuntu-16.04-ppc64le/lastSuccessfulBuild/console

However the job for building Swift 5.1 toolchain with entire test suite execution fails:-
https://ci-external.swift.org/job/oss-swift-5.1-RA-linux-test-ubuntu-16.04-ppc64le/lastFailedBuild/console

This failure is seen due to:
Failing Tests (3):
lldb-Suite :: lang/swift/metatype/TestSwiftMetatype.py
lldb-Suite :: lang/swift/swift_reference_counting/TestSwiftReferenceCount.py
lldb-Suite :: lang/swift/zero_size_generic_self/TestSwiftZeroSizeGenericSelf.py

Still trying to figure this out, but in the interest of time/efforts, any suggestions are welcome.

Comment by Sarvesh Tamba (JIRA)

@adrian-prantl
https://ci-external.swift.org/job/oss-swift-5.1-RA-linux-test-ubuntu-16.04-ppc64le/4499/console

Comment by Sarvesh Tamba (JIRA)

Hi @adrian-prantl ,

The above issues are still in the recent build - https://ci-external.swift.org/view/swift-5.1-branch/job/oss-swift-5.1-RA-linux-test-ubuntu-16.04-ppc64le/4587/

Any help here would be greatly appreciated. Hoping to resolve this one soon.

Without access to a machine that reproduces the issue this is difficult to say from afar.

What I would do is:

1. Try to reproduce the failure from the LLDB command line (by executing actions in the lldb console that reflect what the .py in the test directory is file is doing).
2. Turn this into a standalone command line like `lldb -o "cmd1" -o "cmd2" ... .../a.out`
3. Do the same thing on an x86_64 machine with the same version of lldb to see how lldb is supposed to behave.
4. attach lldb to both lldb invocations and step through it to see where good and bad version diverge
5. Come back here with your findings and we'll try to figure out what's going on.
You can probably speed up step 4 by dumping type and expr log of the good and bad version first and see where they diverge; this will tell you where to put the first breakpoint.