On Mon, Jul 19, 2021 at 05:42:59AM -0500, Felipe Contreras wrote:
> Jeff King wrote:
> > On Mon, Jul 19, 2021 at 02:53:30AM -0500, Felipe Contreras wrote:
> > 
> > > Patrick Steinhardt wrote:
> > > > On Tue, Jul 13, 2021 at 06:07:12PM -0700, Junio C Hamano wrote:
> > > > [snip]
> > > > > * ps/perf-with-separate-output-directory (2021-07-02) 1 commit
> > > > >  - perf: fix when running with TEST_OUTPUT_DIRECTORY
> > > > > 
> > > > >  Test update.
> > > > > 
> > > > >  What's the status of this one?
> > > > 
> > > > From my point of view this is ready, but it's still missing reviews so
> > > > far. The lack of interest seems to indicate that nobody has hit the
> > > > issue so far, and I wonder why that is. Am I the only one who sets
> > > > TEST_OUTPUT_DIRECTORY to a tmpfs directory in his config.mak to speed up
> > > > tests?
> > > 
> > > No, I do the same, and this other fix for TEST_OUTPUT_DIRECTORY is being
> > > ignored even harder:
> > > 
> > > https://lore.kernel.org/git/20210707030709.3134859-1-felipe.contreras@gmail.com/
> > 
> > Note that the linked patch will break Patrick's, because we would no
> > longer set TEST_OUTPUT_DIRECTORY in GIT-BUILD-OPTIONS.
> 
> Is Patrick's patch specific to GIT-BUILD-OPTIONS? Or can
> TEST_OUTPUT_DIRECTORY be taken from the environment?
> 
> Either way I think the priority should be to make the standard tests
> work with TEST_OUTPUT_DIRECTORY.

Right now, TEST_OUTPUT_DIRECTORY as defined in config.mak will be
written into GIT-BUILD-OPTIONS, which then gets sourced by test-lib.sh
automatically. So even if you execute tests directly and not via the
Makefile, it'll still pick up the correct output directory you've
specified in your config.mak. It would likely be possible to take it
from the environment, but that makes the calling interface much more
fragile given that you'll have to remember to always set the envvar.

Patrick