On Wed, 14 May 2008 12:39:11 -0300, Miguel Filho miguel.filho@gmail.com wrote:
On Wed, May 14, 2008 at 11:39 AM, David Ripton dripton@ripton.net wrote:
This means that Python found an illegal virtual machine instruction.
You can cause this by running or importing a .pyc file from a different version of Python, without having a matching .py file to fall back on.
Hmm, that stills look strange, because I didn't change anything.
Encountering this after the process has been running for a long time suggests a memory corruption issue to me. The fact that it happened only once while the process is still running and correctly servicing events further suggests that the corruption was somehow transient, a somewhat unusual occurance to say the least. :)
Before the SystemError was raised, the process probably wrote a tiny amount of debug information to stderr. Is there any chance that got logged somewhere useful? In the normal deamonization mode of twistd it probably would have been discarded since as part of daemonization stdin, stdout, and stderr are all closed. The information is barely useful, though, so it isn't a huge loss, but at least it would point out which code object had the bad opcode. The traceback suggests it was twisted.internet.selectreactor.SelectReactor.doSelect (at least, if you're using select reactor). You could attach gdb and check the the opcodes in the code object for that function, but I imagine they are correct now, since the process is still running.
So, I have no idea what happened, but it sounds interesting. :)
Jean-Paul