[PATCH] D14706: [lit] Implement support of per test timeout in lit.

Chris Matthews via llvm-commits llvm-commits at lists.llvm.org
Thu Nov 19 17:43:55 PST 2015


Ah, I have already forgotten how it works.   Yes, as long as failure is useful, I’m happy!

> On Nov 18, 2015, at 1:22 PM, Dan Liew <dan at su-root.co.uk> wrote:
> 
> On 18 November 2015 at 18:53, Chris Matthews <chris.matthews at apple.com> wrote:
>> Sorry, I was using some silly terminology.  To be compatible the Jenkins and JUnit the xunit reports use the XML field “stacktrace” as the what-went-wrong message.  So that is the part of the report that should say there was a timeout.
> 
> That's not what lit does currently. Lit current dumps information
> inside a ``failure`` tag. Here's some example output
> 
> ```
> <?xml version="1.0" encoding="UTF-8" ?>
> <testsuites>
> <testsuite name='Symbooglix driver' tests='1' failures='1'>
> <testcase classname='Symbooglix driver.Symbooglix driver'
> name='slow.bpl' time='1.01'>
> <failure >
> Script:
> --
> 
> --
> Exit Code: 1
> 
> Command Output (stdout):
> --
> Command 0: "sleep" "1"
> Command 0 Result: 0
> Command 0 Output:
> 
> Command 0 Stderr:
> 
> Command 1: "false"
> Command 1 Result: 1
> Command 1 Output:
> 
> Command 1 Stderr:
> --
> 
> </failure>
> </testcase>
> </testsuite>
> </testsuites>
> ```
> 
> My patch to lit doesn't give very useful output when a timeout is hit
> which I need to fix.
>> 
> 
>> What I would really love is some help for the developer.  Blah timed out is a bad scenario, it basically means “Im hung, but I won’t tell you where”, now go and try and figure that out our reproduce me somewhere else.  On Darwin (and I assume Linux too?) there are tools like spindump to give you a stack shot of a hanging process.  If we could do something like that, that would be wonderful. Then you get a real stack trace of your hang, right in the test output.  Since hangs are so hard to reproduce and diagnose, I think this would be a big win.  I don’t know how to do this in a OS agnostic way though?
> 
> It might possible to do something in the future. On Linux you could
> attach lldb or gdb and print a stacktrace for every thread. I think
> that's out of scope for this patch though.
> 
> Thanks,
> Dan.



More information about the llvm-commits mailing list