I was going to argue, but it's not worth it. What you propose is fine.

On Sun, Jan 28, 2018 at 10:03 PM, Nick Coghlan <ncoghlan@gmail.com> wrote:
On 29 January 2018 at 14:43, Guido van Rossum <guido@python.org> wrote:
> So why can't you just run "make test" if that's faster?

I can (and do), but I also run it the other way if I need to pass
additional options. I'll then notice that I forgot -j0, ctrl-C out,
then run it again with -j0.

That's a minor irritation for me, but for folks that don't already
know about the -j0 option, they're more likely to just go "CPython's
test suite is annoyingly slow".

To provide a bit more detail on what I'd suggest we do:

* "-j1" would explicitly turn off multiprocessing
* "-j0" and "-jN" (N >= 2) would explicitly request multiprocessing
and error out if there's a conflicting flag
* not setting the flag would be equivalent to "-j0" by default, but
"-j1" if a conflicting flag was set

The testing options that already explicitly conflict with the
multiprocessing option are:

* -T (tracing)
* -l (leak hunting)

"-j1" would likely also be a better default when the verbosity flags
are set (since the output is incredibly hard to read if you have
multiple verbose tests running in parallel).


Nick Coghlan   |   ncoghlan@gmail.com   |   Brisbane, Australia

--Guido van Rossum (python.org/~guido)