On Wed, Feb 21, 2018 at 08:53:16AM -0800, Junio C Hamano wrote:
> Even though keeping track of list of known-leaky tests may not be so
> interesting, we can still salvage useful pieces from the discussion
> and make them available to developers, e.g. something like
>
> prove --dry --state=failed |
> perl -lne '/^(t[0-9]{4})-.*\.sh$/ and print $1' | sort >$@+
> if cmp >/dev/null $@ $@+; then rm $@+; else mv $@+ $@; fi
>
> could be made into a target to stash away the list of failing tests
> after a test run?
Unfortunately there are some caveats in that snippet:
1. You are using prove.
2. You are using --state=save in the initial run.
I think we might be better off having the test scripts write to
test-results/*.counts even when run under a TAP harness, and then we can
have a consistent way to get the list of failed tests (we already have a
"make failed" that works this way).
-Peff
