BTW regarding profiler caption, you can set it using dbg_set_profile_caption($caption) debugger API. I've added it quite long ago.

Hi Dmitri,
Of course this is a micro difference, but my point was different here: are you summing up 2 positive numbers? Or doing 2 completely different measures? I understood that you sum values on the client. In the first case it doesn’t make sense that if you do A + B = C and both A and B are > 0 you’ve C < A.
In the second case it’s possible, expecially on Windows where timers behave randomly with measures < 70ms

I was just trying to find sonething from a very small and reproducible example

May I ask what problem you are chasing now?
If you want to blame us that we're doing something wrong in profiler calcuations, okay, point taken. But that's probably enough. This 2 microsecond discrepancy has no relation to what you have originally experienced.
If you want to help us identify the problem you encountered in the beginning, you can do this by providing reproducible case showing that problem with discrepancy like fraction of seconds or even seconds, at least several milliseconds, but not 2 microseconds. Unfortunately you refused to.
I'll highly appreciate if you change your mind.

I'm trying to help finding the original issue. I'm hoping that this easily reproducible case can help solving the other bigger case.

I'm not being able to reproduce it again, this is the problem. As you can see I'm continuing to try to help finding the issue.

Of course there is no reason to blame you for anything.

Hi Dmitri,
I've modified the trivial script and I was able to get bigger differences; here the current script:

<?php
for($i=0;$i<10000000;$i++)
md5('abc'); // line 3
?>


and these the results under the "Functions" tab:


Summary
File: C:\Users\pc\Documents\SITI\Shopping\www\html\md5.php
Line number: 3
Line time: 5,962 sec
Line time with children: 4,153 sec
Functions total: 7,235 sec
Page total: 7,237 sec

Now Line time with children is 1,8sec smaller of Line Time.

here the screenshot


Is this a good starting point to debug the problem of "Line time with children" < "Line time" ?

This discrepancy is of 0.181 microsecond per iteration. You think it's too bad?

I’m trying to understand the math behind. At this point I’ve to assume that it’s not true that “time with children” is the function time plus the time spent into the children; right? They are 2 completely independent measurements.