b.Loop(), count=1 by default and 3+ for the numbers quoted here, 200 ms–1 s per benchmark, allocations always reported (-benchmem).- Fixtures are synthetic but shaped like the real thing: prose sections,
lists, tables, links, code fences in five languages, chrome to strip
(
nav,share,footer), and articles that grow past 300 KB. - The network is loopback, so
LoadLoopbackmeasures fetch + convert + allocation, not the internet; a real page adds WAN latency on top. - Cache benchmarks flip the real switch (
cacheOn) and use the realloadCmd, not a stub of it.
performance / methodology
Methodology
b.Loop(), count=1 by default and 3+ for the numbers quoted here, 200 ms–1 s