Re: gprof reporting zero times for one program

[Nick Clifton]

Hi David,

> The only change other than the use of multiple files, instead of
> just one,

Having multiple source files rather than just one definitely should not be 
causing this.

> that I can think of which might be causing this
> is that the new version of nettee calls alarm(0) and alarm(time)
> more frequently than did the previous version. If gprof embeds code
> which uses the same mechanism it might never be seeing its timers going
> off. 

Hmm, well profiling support is implementation specific, so you may be on to 
something here.  What typically happens when profiling is enabled is that the 
compiler inserts a call to a special function called __mcount() which takes 
place during the prologue of every compiled function.  __mcount records which 
function is being called (based on the PC) and the time the call was made, 
based presumably on a timer/clock reading of some sort.  When the program exits 
special code in the exit handler formats and dumps this information into the 
gmon.out file.

So if you are manipulating the timer/clock then maybe this could be the cause 
of the problem.  Can you try suppressing just that code temporarily to see if 
this restores proper profiling ?

If that does not work or cannot be done easily, can you post a gmon.out file 
for us to look at.  (Compressed of course).  I suspect that you have a compiler 
and/or C library problem here rather than a broken gprof which is refusing to 
read a proper gmon.out file, but it never hurts to be sure.

Cheers
  Nick