Re: [PATCHv6 16/16] Introduce portable_unset and use it to ensure proper && chaining

Jonathan Nieder [off-list ref] writes:

Ævar Arnfjörð Bjarmason wrote:

quoted

                        I don't know if we came to an agreement on
using test_might_fail for non-builtins

Non-builtins: okay.
Non-git commands: no.

That's just my take, though; if the project consensus is different,
who am I to resist it?

Hmm, the pros and cons between allowing test_might_fail on normal commands
vs introducing cmd_might_fail are questionable, though.  I cannot think of
a case where we would want to tolerate death by signal or segv on non git
commands by using cmd_might_fail.  At least I do not think that is a
common thing we would want to ignore, and in a rare case, it probably is a
good idea to say that explicitly with something like...

	{
        	frotz
                : the above might randomly segfault but we do not care
	}

Junio C Hamano wrote:

Hmm, the pros and cons between allowing test_might_fail on normal commands
vs introducing cmd_might_fail are questionable, though.  I cannot think of
a case where we would want to tolerate death by signal or segv on non git
commands

Okay, makes sense.  I would be happier if exit statuses 129 and 129-256
were disallowed by test_must_fail, too, but that does not make it any
less useful for cases like "test_might_fail grep needle haystack >result".
And it is better to avoid introducing yet another command to learn.