Just an idea. Currently the timeit module in command line automatically choice most appropriate time unit for human readable output. But this makes the output less machine readable. I propose to add new option which specifies fixed time unit (usec, msec, sec).

If you want the output to be machine readable, better go the full way and add a --json option, IMO.

My 2 cents - I'd split timeit.main function into processing part and argument parsing + data printing part so that you could use Python to call the first one to get the actual results (timings, numbers of loops etc.) and then present the data to the outside world in any needed format - no text parsing needed.

I created a patch that allows users to specify a time unit output using the "-u" flag. Example usage:

python -m timeit -u "msec" '"-".join(str(n) for n in range(100))'
>> 10000 loops, best of 3: 0.156 msec per loop

python -m timeit -u "sec" '"-".join(str(n) for n in range(100))'
>> 10000 loops, best of 3: 0.000159 sec per loop

You can specify msec, sec, or usec. I am also interested in having a JSON option as mentioned by Antoine. Any thoughts?

Updated patch to include unit tests.

Just wanted to check to see if there was anything else I should do regarding this issue.

Implemented proposed changes to patch. Simplified "for-loop" and implemented invalid input test.

Incorporated updates suggested by serhiy.storchaka

It will be better to write error message to sys.stderr (see test_main_exception).

Updated patch to log to stderr.

Added newline after error message.

The patch is doubled.

Whoops. Sorry about that. Super embarrassing.

Re-uploaded the patch to remove duplication.

Besides too long line with sys.stderr.write last patch LGTM.

Ah, I forgot. This feature should be documented in Doc/library/timeit.rst.

Updated documentation and fixed a few lines that were too long.

Just wanted to check in and see if the documentation change I made is sufficient. I tried to copy the other entries for the command line arguments. Let me know if I can improve it in any way.

LGTM.

What needs to be done to close this one out?

This feature is for Python 3.5. I think, we will not commit anything to Python 3.5 until Python 3.4 branch is created (in other world, released). So, be patient. :)

The branch appears to exist now.

Anything else need to be done on this patch?

Anything else need to be done on this patch?

I've added a couple of comments on Rietveld.

New changeset 1ebf8d5b7d60 by Robert Collins in branch 'default':
Issue #18983: Allow selection of output units in timeit.
https://hg.python.org/cpython/rev/1ebf8d5b7d60

New changeset ed34dd00405e by Robert Collins in branch 'default':
Fix patch attribution for issue 18983.
https://hg.python.org/cpython/rev/ed34dd00405e