Re: [LAU] [Zynaddsubfx-user] zyn and the art of software maintenance

On Thu, 17 Sep 2009, Grammostola Rosea wrote: When I was tracing 'zombification' issues in Hydrogen, it was very helpful to me (the developer) to know: * jackd settings when firing up the application (e.g. jackd -R -d alsa -p 4 -n 2 -r 192000 -Xseq) * A little more about the system (memory, CPU, distro, jackd version) * Any command-line options you used to start the application and cause the fault. Obviously, Zombies happen more often with aggressive JACK settings. On the other hand, even with that info it is generally hard to trace out a Zombification issue. The zombie notification happens a few milliseconds _after_ the event that waited -- so, backtraces weren't very helpful. Increasing your logging can corrupt your measurements because I/O is one source of a zombie. It was even possible, for instance, for one application to take just a little too long (leaving no time for the 2nd app). JACK would then blame the zombification on the 2nd app. I ended up having to do a static code analysis to search for memory allocation, re-do our logging scheme (because its mutex was causing locks), and discover some new taboo functions. The biggest surprise (for me) is that Qt's QString constructor and operator= could not be used in real-time code.