On Tuesday 09 September 2003 15:09, Ian Reinhart Geiser wrote:
> [...]
> 	The idea is that a stat and compair of QDateTime is faster than the
> reparse.

You have to be careful with this assumption: calltree only simulates user 
level, so you don't get any info on how much time the stat() took. If 
possible, a real measurement should be done, also.
But here, I suppose that reparsing includes reading the file, so simple time 
checking is always better.

Regarding the backtrace code:
You can request this info from calltree: Calltree can output Call-chains by
concatenating symbol names, separated by an apostroph. I.e. the call chain A > 
B > C gets "A'B'C".
To not only see the function C in the profile dump, but all call-chains of 
length 5 ending in C, use "calltree --fn-caller4=C ...".
The call-chains will appear to KCachegrind as separate functions themselves, 
i.e. instead of C you will get perhaps 50 functions ending in "...'C".
Remark: I hope this feature works as announced ;-)

It's on my TODO for KCachegrind to represent call-chains in a nice way. But as
nobody used this feature till now, it's not important.

Note that you can request to globally get e.g. call-chains of length 3 by 
using "calltree --fn-caller=3". This will enlarge the dump, and you get 
around 10x the number of functions in KCachegrind (the call-chains), so
this is only useful when handled by KCachegrind in a nice way.

Josef