I finally figured out why bash has its own “time” the other day. It’s documented, but poorly / subtly:

$ help time

time: time [-p] pipeline
    Report time consumed by pipeline's execution.

It’s a parsing issue. The following command times a whole pipeline, which ends up being the length of the longest command, because the command are started in parallel.

$ time sleep 0.1 | sleep 0.2

real    0m0.202s
user    0m0.002s
sys     0m0.001s

This is a pipeline where the first command is /usr/bin/time, so the time is only reporting 0.1 seconds.

$ /usr/bin/time sleep 0.1 | sleep 0.2
0.00user 0.00system 0:00.10elapsed 0%CPU (0avgtext+0avgdata 1812maxresident)k
0inputs+0outputs (0major+78minor)pagefaults 0swaps

Likewise you can do this:

$ time {
> sleep .1
> sleep .2
> }

real    0m0.302s
user    0m0.000s
sys     0m0.001s

This won’t work with /usr/bin/time because it will try to run the command {. I probably should copy this into a blog post: http://www.oilshell.org/blog/

Perhaps worth mentioning that this article seems to be about the time(1) from BSD (presumably from opensource.apple.com/source/shell_cmds/shell_cmds-198/time/ specifically) whereas if you’re running Linux you probably have time(1) from savannah.gnu.org/projects/time. The GNU version uses -v instead of -l.