Add a built-in runtime profiler

[TrixNEW]

Summary

zphp should include a profiler for finding expensive PHP code and runtime bottlenecks.

For example:

zphp profile app.php

or:

zphp profile serve app.php

Useful metrics

Track:

• function execution time
• call counts
• self time vs total time
• allocations
• allocated bytes
• peak memory
• GC activity
• VM/opcode execution
• blocking I/O time
• worker/thread usage
• Fiber scheduling time

Example output:

Function                 Calls     Self       Total
World::tick               2000     420ms      780ms
encodeChunk                800     300ms      310ms
Player::tick             50000     250ms      400ms

Optional output formats could include:

• text
• JSON
• Chrome Trace
• flamegraph-compatible data

Profiling should have minimal overhead when disabled.

---

[TrixNEW]

Its just important this doesnt cause any mem leaks or takes away performance

[nvms]

I think this is generally a good idea. Your example output infers a profiler that hooks every function call, but your follow-up comment says it cannot cost performance (which I agree with), but these two ideas fight eachother. A hook on every call means lots of work in the VM hot path, slowing everything down whether or not anything is being profiled.

What should probably be done instead is a sampling profiler that fires every few ms and records what the VM is currently executing by walking its frame stack, and when it is off the only cost is checking the "is profiling enabled" flag. When on, overhead should be minimal. Output can drive a flamegraph. Downside is that call counts come out as estimates instead of exact numbers, but I honestly think that's a good tradeoff.

I disagree with anything that costs performance even when disabled.

I will keep thinking about this a bit more and get to it soon.