Release Notes

0.4.1

Make warmup and count estimation more robust Also splits out debug and warning messages.

Will now warn if compilation occurs during execution count estimation.
Compute and subtract the measurement overhead Estimate the measurement overhead on first run, and then subtract it from subsequent estimates.

An explicit overhead can be supplied with the :overhead keyword.
Pass a map instead of varargs In all but the top level bench and quick-bench macros, use a map for options, rather than varargs.
Make the final gc run immediately after sampling Avoid doing any other memory allocation after the end of sample collection.
Make warmup-for-jit more efficient Accumulates runtime in chunks, rather than timing individual calls.
Use a mutable field for timing loop storage Replace the use of an array for the timing loop storage, and remove the use of a reducer. Testing showed no difference in the resulting measurements, and this simplifies the code.
Add benchmarks for timing loop implementations In order to quantify the impact of several timing loop implementations, adds benchmarks to quantify them. There seems to be no difference in the result between using an array, or a mutable place.
Use mu for microsecond units. [#14]

Add macro 'benchmark-round-robin' that is like benchmark, but takes a sequence of expressions The goal is to see whether the results are noticeably different when executed 'round robin' instead of doing the same expression over and over for an extended period of time (e.g. 1 min). Instead try 10 different expressions for 100 millisec each, for example, and then repeat that 60 times if :sample-count is 60.

One needs to be cautious in interpreting the results, e.g. if one expression in the sequence tends to leave behind lots of garbage that won't be collected until a later expression is being executed.
Change execute-expr so it has less overhead per execution of the benchmarked expression.
Return options used as part of benchmark* return value
Add Clojure version and system property sun.arch.data.model to runtime-details return value.
Add the values of system properties java.version and java.runtime.version to return value of runtime-details.

Remove unused code. Makes criterium work on clojure 1.5.0-master-SNAPSHOT. Fixes #11.
Add :gc-before-sample option Passing this option as true will force gc before every sample is taken.
Fix labelling of quantiles, and drop display of intervals on the estimator distributions
Force the collection of samples before calculating any statistics
Remove compilation and reflection warnings
Add tests for bootstrap-estimate and bootstrap-bca

Use doubles in time calculations to prevent overflows
Switch to use function, add :reduce-with, and output CI on mean This gives more consistent results
Update to clojure 1.3.0, and project to 0.2.0