[Lldb-commits] [lldb] r219102 - Call SBDebugger::Initialize/Terminate from within Create/Destroy.

[Ed Maste]

On 7 October 2014 02:18, Matthew Gardiner <mg11 at csr.com> wrote:
> Hi Ed,
>
> Which test was the first to segfault in the suite. I've looked at the stdio
> page from the check-lldb target from the attached link, and it looks like
> it's this statement which indicates the bot failure:
>
> FAIL: LLDB (suite) :: TestFormattersSBAPI.py (FreeBSD
> llvm-amd64.freebsd.your.org 10.1-STABLE FreeBSD 10.1-STABLE #0 r272281M: Mon
> Sep 29 14:19:16 UTC 2014
> root at llvm-amd64.freebsd.your.org:/data/obj/usr/src/sys/GENERIC amd64 amd64)

Yes, that's the first failure. Unfortunately it seems the segfault is
hidden by the test suite, so it's not that obvious from the log where
it happens.

I reproduce it with:
python dotest.py --executable /.path.to./lldb -C /usr/bin/clang -v -t
-p TestFormattersSBAPI

The Linux and OS X thread libraries are more forgiving of invalid use
than FreeBSD's, which may explain differences in the test results.
For example when I started working on LLDB there were a few instances
of calling _unlock() on an unlocked mutex, or from a thread other than
that which called _lock().  That seemed to work on OS X and Linux, but
crashed on FreeBSD.

> I'm keen to know as I'd like to set up a freebsd system to replicate this
> type of failure.

Excellent, let me know if you need anything.  I have some instructions
for setting up a FreeBSD VM and building LLDB here:
https://wiki.freebsd.org/lldb