Re: JIT compiler changeset results in plot failure with segfault

[Michael Goffioul]

On Sat, Nov 2, 2013 at 11:43 AM, John W. Eaton <address@hidden> wrote:

On 11/02/2013 10:43 AM, Michael Goffioul wrote:

I guess not, except if you have a memory corruption somewhere that
messes up with the gdb stack. Maybe running it into valgrind can reveal
something. I also noticed 2 different versions of LLVM in the backtrace:
frame #2 uses libLLVM-3.2.so.1, while frames #0 and #1
use libLLVM-3.0.so.1. Maybe this is normal, but I figured I mentioned it.

Oh, that can't be right.  I removed the 3.0 packages and built again
from scratch and I no longer see the crash.  Thanks for catching
that.

I still think that we shouldn't be using or initializing llvm unless
the JIT compiler is enabled, so I'll try to track down where that is
happening.

From the look of it, it looks like initialization code executed when a library is loaded by the dynamic linker (equivalent to the DllMain in the Windows world). So you might not have much of a choice than having LLVM initialized.

Michael.