benchmark: always throw the same Error instance

[addaleax]

Stack trace capturing currently accounts for 40 % of the benchmark
running time. Always throwing the same exception object removes
that overhead and lets the benchmark be more focused on what it is
supposed to measure.

Refs: #34512 (comment)

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

[jasnell]

Hmm, I'm torn on this. While I get the reason for the change, it makes the benchmark even more unrealistic compared to real world user code. Our microbenchmarks are already overly optimized as it is. Not a blocking concern, however.

[addaleax]

@jasnell I mean, that really depends on what we’re measuring here, right? The point is to measure the performance effect of async_hooks on Promises. Let’s leave that out which isn’t relevant to it.

(For example, the other async_hooks benchmarks are very diluted because they’re full-blown HTTP servers – that might be closer to a real-world example, but it makes it harder to measure the perf impact of async_hooks themselves. For profiling, they are basically useless and I always needed to go to the MessagePort benchmarks instead.)

[jasnell]

Yeah like I said, I'm torn on it and don't consider it blocking. I just think we way over index on microbenchmarks in general.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/32518/

[puzpuzpuz]

Landed in b14ce72