The current profiler averages all calls to a function together, regardless of where or in what context the function was called. This often makes it hard to draw useful conclusions from the data, especially when a function behaves differently depending on how it's used.
I think the profiler would be much more useful with the following improvements:
-
Instead of just listing functions, show the data in a tree structure. If you select a function, you should be able to see all the functions it calls, how long each call took, and drill down further into each of those functions.
-
This would let developers see how functions perform in context, rather than just as a flat average.
-
As a simpler alternative, the profiler could group and display performance data per call stack. Right now, it shows different call stacks but doesn't show how much time the function took in each of them. Adding timing info here would already be a big step forward.
These changes would make it easier to spot bottlenecks and understand performance in real-world usage, rather than trying to interpret averaged data without context.
