mark() columns memory, result and mem_alloc columns are now always included, even if they are unused.New bench_process_memory() function, to return the current and maximum memory used by the current process. This uses system functions to track memory, so can measure memory outside of R’s GC heap.
New workout_expressions() function, a low-level function to workout a list of expressions, like those obtained via parse() from a file.
mark() gains a memory argument to control if it records memory allocations, set memory = FALSE to disable recording memory allocations, which can be helpful when trying to benchmark long pieces of code with many allocations (#62).
mark() now permits empty arguments, e.g. accidental trailing commas (#61).
mark() now errors correctly when the expressions deparsed length is different.
bench_expr objects now work better with the upcoming versions of tibble and vctrs (@romainfrancois, #64)
autoplot.bench_mark() provides a more informative error if the ggbeeswarm package is not installed (@coatless, #69).
Update documentation of bench_mark columns (@jdblischak, #67).
bench_memory() examples no longer fail if they are run with R that does not have memory profiling capability (#56).
bench_expr now has a class of c("bench_expr", "list") rather than c("bench_expr", "expression"), as it is really a list of calls rather than a true expression object. (https://github.com/r-lib/vctrs/issues/559)
summary.bench_mark() gains a time_unit argument, so you can report all times in a consistent scale if desired (#18, #26).
bench_mark() now checks for user interrupts, to allow you to stop benchmarking if it takes longer than you were expecting (#49).
New bench_memory() to capture just the memory allocated by an expression.
bench_time() is now an alias for system_time().
unnest.bench_mark() is now compatible with the upcoming tidyr 1.0.0 (#48, #51)
New hires_time() allows you to explicitly capture high resolution time points.
workout() a new function which makes timing multiple expressions in turn simpler.
mark() now internally uses a tempfile rather than a textConnection, as the latter has a 100,000 character limit on some platforms (#27)
mark() no longer returns the mean or max values and the column order has been tweaked to try and put the most interesting columns first (#37)
Errors during evaluation now halt execution (#28, #43)
scale_bench_time() and scale_bench_bytes() now allow you to use a non-logarithmic scale.
NEWS.md file to track changes to the package.