Not sure. Another reason I like these types of benchmark…even with all their “flaws”, if the “flaws” are consistent, they can catch regressions.
This commit changed the data update test:
https://github.com/TechEmpower/FrameworkBenchmarks/commit/96deafcf351ae0b7ae44ae51b292cfa829eab382#diff-92a2750f03cbfd63d2c3b27eeeef442a
You can see it no longer serializes through Poison’s @derive attribute which, according to the Poison docs, should be used for “maximum performance”.
There was also a more recent commit that changed many (faster) encode_to_dataio! to encode! which isn’t going to help anything.
