FYI, since there was never any action on this, I'm closing a related JRuby bug: http://jira.codehaus.org/browse/JRUBY-5381 If this is revisited in the future, we'll be happy to follow ruby-core's lead. - Charlie On Tue, Jan 18, 2011 at 3:18 PM, Charles Oliver Nutter <headius / headius.com> wrote: > On Tue, Jan 18, 2011 at 2:24 PM, Evan Phoenix <evan / fallingsnow.net> wrote: >> The second diff also changes benchmark.rb to also use tutime out of Struct::Tms instead of utime. This provides more relevant timing of the block than using utime. > > I think that's debatable. The times listed in the Benchmark results > have always been full-process user and system times. Changing them now > would skew benchmark numbers on systems where some work is happening > in concurrent threads. On implementations like JRuby and Rubinius, it > might exclude GC and JIT overhead entirely, which may be important > numbers for a given benchmark. > > Perhaps instead, Benchmark should report all timings (in a readable > format when printed) rather than leaving users to guess what the > numbers mean (and guess wrong once it switches to thread timings). > > The addition of thread timings to Struct::Tms seems good to me. > > - Charlie