Can the allocator passed to `init()` be used to track allocations of tests?

[inge4pres]

Thanks for making zBench! 🙏🏼

While exploring the library for the first time, I noticed that tracking allocation is possible only by making the code under test consume the allocator passed to run().

I can imagine why this is designed this way, and I was curious to know if you think it poses a limitation.

What is the suggested approach to benchmark a specific function/method that does not consume an allocator directly, but rather inherits the allocator from a struct?
As a reference, in this benchmark test I would like only to record allocations for the histogram.record() method call.
The problem is, record() does not accept an allocator, so apparently I am have to run the entire setup in the benchmarks too.

Is there a better way?

[hendriknielaender]

You are correct, it cannot. This is a deliberate design choice.

For example, it's used to create the ArrayList(Definition) that stores all your benchmark functions. If we used this same allocator for tracking, the framework's own internal allocations (like adding a new benchmark definition) would be mixed in with the code you're actually trying to measure. This would pollute the results.

A possible solution could be to give the Histogram an allocator that you can control from the outside. The best way to do this is with a Swappable Allocator. There you would need to create a custom allocator that acts as a proxy, forwarding all its calls to an underlying allocator that you can change at runtime.

Maybe we could think of adding this custom allocator.