This may be a good explanation. When I did run the tests, the result was dependent on the number of tests executed. E.g. when only running ComplexTests, no failures were reported. Only when the FastMathTests were also executed.
As the FastMathTests do a lot of computations, it is safe to assume that JIT jumped in, while for the ComplexTests alone, the execution might have completely been run in interpreter mode. It still does not explain why this only happens on specific machines, and also only under specific conditions. Thomas On Wed, Jan 14, 2015 at 12:26 PM, sebb <seb...@gmail.com> wrote: > On 14 January 2015 at 00:36, Phil Steitz <phil.ste...@gmail.com> wrote: > > Why only sometimes? > > > > I'm not sure. > Is a test run completely deterministic, or can the order of tests > vary? i.e. could the JIT see different conditions in different runs > and sometimes decide not to optimise away the code? > Or is the JIT perhaps affected by other activity on the box - e.g. > does elapsed time affect optiimisation? > > If there were some conditions that occasionally prevented > optimisation, it could explain the results we are getting. > > Whatever the cause, I think I am right in saying that the failures > have not occurred with 1.6+ > > > > >> On Jan 13, 2015, at 4:36 PM, sebb <seb...@gmail.com> wrote: > >> > >> https://builds.apache.org/job/Commons%20Math%20H10/38/console > >> H11 - java6 - OK > >> > >> https://builds.apache.org/job/Commons%20Math%20H10/39/console > >> H11 - java7 - OK > >> > >> https://builds.apache.org/job/Commons%20Math%20H10/40/console > >> H11 - java5 - failed > >> > >> https://builds.apache.org/job/Commons%20Math%20H10/41/console > >> H11 - java5 with -Djava.compiler=NONE - wait for it - a bit longer > >> > >> > >> keep waiting > >> > >> > >> > >> and waiting > >> > >> > >> (disabling optimisation makes the test MUCH slower) > >> > >> > >> > >> still waiting? > >> > >> > >> > >> a bit more ... > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> SUCCESS! > >> > >> So I think it's almost certainly a JIT optimisation bug in 1.5 > >> > >> QED? > >> > >> --------------------------------------------------------------------- > >> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org > >> For additional commands, e-mail: dev-h...@commons.apache.org > >> > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org > > For additional commands, e-mail: dev-h...@commons.apache.org > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org > For additional commands, e-mail: dev-h...@commons.apache.org > >
