Can the time split by nest level?
So xdebug only sum up the total taken of the current nest level?

Currently my XDebug output here goes crazy, WinCacheGrind telling me percentage values in the 10000% range... (heavily recursion together with Smarty3) (That was actually bug 0000357

A slightly modified testcase (less recursive, more time per call):
<?php

function c() {
usleep(1000);
d();
}
function d() {
static $x;
if (!isset($x)) $x = 0;
$x++;
while ($x <30) c();
}

c();

?>

Can confirm this issue.
This, of course, also applies to object methods.
Really hard to realize that there's something wrong because if the program contains more than just nested functions then you won't realize that the cumulative is actually wrong.
It just seemed way too high, which got me looking into the issue ...

I desperately need a fix to this, or I will have to go back to ZendDebugger... working on a project where recursion is unavoidable as its goal is to wrap together pieces of self-referencing data.

So, here's my 2 cents:
http://www.exteon.ro/ro/produse/unelte-php/xt2callgrind
It's tested with a fairly large script of 100K+ function calls, lots of recursion, results are excellent.

It's a very large computational effort though, I now doubt it can be implemented in real time, might thou be translated in a C post-processing feature.

I do have one related question though, if it's in the Docs I missed it: with the profile or trace trigger, I get timings 200% (for large scripts) to 600% (for small scripts) of the real execution time. Do you have some stats linking real execution time, traced execution time and call overhead? Or what is expected of? Should the overhead be roughly linear with the execution time, or roughly linear with the # of function calls?

Any plan on fixing this bug?

It's still wrongly reporting time spent in recursive function in xdebug 2.3.3 even if we push detect cycles...

I generated from vallgrind for a C program the equivalent example given here and kcachegrind show something relevant if we use detect cycles.