Inquisitor

We have the same algorithm implemented in several places: for some tests we do benchmarking using a "dumb" wall clock timing, using time command.

A proposal is to make a "timed_run" library function that does the following:

• gets the target command (with full options, etc) to measure in $@
• runs this command under time, redirects its stdout and stderr in $DEBUG_TTY
• checks exit code — fails current test if there was an error
• gathers the data from time
• cuts wall clock from them and outputs it in stdout
• does proper garbage collecting (i.e. deletes temporary files, etc)

This way, most of encoding tests, for example, could be reduced to simple one call:

benchmark_submit_float \
    'Encoding time' \
    "`timed_run flac -$COMPRESSION_LEVEL -f -o /dev/null $SOURCE`" \
    sec