12 октября 2011, 19:14 от Yves-Alexis Perez <cor...@debian.org>: > > How exactly did you cook this?
I don't understand exactly what that means, so I try to answer to my understanding of your question: "how exactly did you find this?". gprof found that most cpu time was spent around 'check_preedit' I placed checkpoints (printfs) over code and found that after pressing Enter key event reissued many times, that caused high cpu load. Then compared code of mousepad with leafpad's one and noticed that leafpad passes NULL instead of pointer to string, so I commented out that piece of code in mousepad too. Assumed strlen(str) must be non-zero and changed default return value to TRUE (that was wrong). That caused immediate return after testing check_preedit() return value. After uncommenting code in check_preedit I forgot to remove "return TRUE;" line. And that was the only change in mousepad's code then. Suprisingly it worked just as I want. That wrong return statement caused part of code to be jumped over, so I tried to understand what that code does and noticed those &=. Variable `event' is passed by reference so changing its content apparently a mistake I thought. Something like this.
