Comment by dan-robertson
2 days ago
I kind think metrics are not the answer and instead one needs taste. Obviously performance is multidimensional both in what one measures (latency vs throughput) and as a function of the input. The solution you imagine that is slightly faster in the test could avoid (or introduce) different worst-case or asymptotic behaviour, for example.
I argue we shouldn’t be doing this at all; but, if we have to do to whatever insanely arbitrary metric a project/product/eng leader wants, this is probably a better metric than code length.