[apologies for the off-list mail] Thanks. For another kind of robustness, is it really necessary to throw away all the benchmark results when one (translate) fails? My run without --fast just fell over after well over a day of running :( Cheers, mwh On 1 May 2013 21:16, Armin Rigo <arigo@tunes.org> wrote:
Hi Michael,
On Wed, May 1, 2013 at 3:54 AM, Michael Hudson-Doyle <micahel@gmail.com> wrote:
File "/home/mwhudson/src/benchmarks/unladen_swallow/perf.py", line 170, in TScore return (avg(sample1) - avg(sample2)) / math.sqrt(error * 2) ZeroDivisionError: float division by zero
The code could be more robust indeed :-( It's because hexiom2 runs only once in --fast mode, so it ends up with error == exactly 0.0. Fixed in f7abffc04667.
A bientôt,
Armin.