tsp42 Wrote:The leak should be closed now.
Remarkable example of how well written the libcmyth code and documentation is :-)

Code:

/**
* Atomically decrement a reference count variable.
* \param valp address of atomic variable
* \return incremented reference count
*/
static inline unsigned
__mvp_atomic_decrement(mvp_atomic_t *valp)
{
    mvp_atomic_t __val;
#if defined __i486__ || defined __i586__ || defined __i686__
    __asm__ __volatile__(
        "lock xaddl %0, (%1);"
        "     inc   %0;"
        : "=r" (__val)

dubstar_04 Wrote:Wow,

What an improvement!!!

It starts up in seconds and memory usage is about 50MB now instead of 1.5GB!!

One of the cpu cores is still pegged at 100% while xbmc is running, is there anyway of finding out what might cause it? maybe feedback to dushmaniac?

thanks again,

Dubstar_04

tsp42 Wrote:The high CPU utilization on linux should be fixed now. Using a profiler is usually the best method to finding the cause. I use intel vTune on linux.

tsp42 Wrote:Maybe related to this?
I will try to improve the timers (and maybe the sorting of records) before investigation this leak further (not as severe as the last leak). It is a bit complicated to map the MythTV timers to the XBMC PVR timers (suggestions are appreciated).
I've been thinking about showing a custom window when adding a recording with the different mythtv specific options at least this would reduce the need to mess around with the XBMC PVR code.

tsp42 Wrote:I've added sorting of recordings by date to the context menu.

dubstar_04: Can you check if the missing timers are in the sql database (in the record table)?