On Wed, Apr 29, 2020 at 03:10:50PM -0700, Junio C Hamano wrote:

Taylor Blau [off-list ref] writes:

quoted

To me this seems a little overkill, but it may not be on environments
where an extra subshell incurred by 'test_might_fail' might be overly
expensive.

It comes from the same principle as "we are not in the business of
catching segv from system tools---don't use test_must_fail on
non-git commands".  Adopting the convention happened quite some time
ago and that was why I checked if we failed to document it.

What I wondered was if it is overkill to document the convention; if
the convention was overkill is not a question at this point.

Ah, fair enough. Thanks for a patient explanation.

Thanks,
Taylor