On Fri, Jan 19, 2024, at 20:57, Ihor Radchenko wrote:
> May you please redo the benchmarks using
> 1. M-x profiler-start <RET> cpu <RET>
> 2. Run agenda
> 3. M-x profiler-report
> 4. M-x profiler-report-write-profile <RET> profile.el <RET>
> 5. Share two saved profiles, repeating the above steps with newer and
>    older Org mode.

Please find attached.

> ELP benchmarks on the whole Org mode are useless - they create huge
> overheads on frequently called fast functions, completely messing up the
> information.

I found the profiler-reports harder to read because of the (native-compiled?) lambda that show up and harder to compare since the metric is relative with no absolute reference but well noted and will remember.

-- 
Alexandre Avanian