Re: Fw: [Lilypond-auto] Issue 2547 in lilypond: Fix documentation of making footnotes work via tweak.

[John Mandereau]

Il giorno gio, 30/08/2012 alle 12.52 +0100, Graham Percival ha scritto:
> On Thu, Aug 30, 2012 at 11:38:51AM +0200, John Mandereau wrote:
> > every new comment on those issues with old patches will trigger a test.
>
> That's definitely overkill!  What if I post a comment saying "yes,
> this patch definitely looks bad"?

This is not only overkill, this is of course utterly maoingly broken and
not intended, I was simply pointing out this behaviour, without
approving it.  I'll see ASAP how to check the date of last patch.


> We could, but I think there's a difference between people who work
> slow/infrequently vs. abandoned patches.  I mean, Mike's skyline
> patch and Ian's guile 2.0 work have probably both seen periods of
> not being changed for 2 months, but both are still being worked
> on.
>
> I don't think that we should automatically declare patches to be
> abandoned.

A different story goes for large changes like Mike's skyline patch and
Ian's guile 2.0 work and for smaller contributions, or contributions;
for the former, I agree that the issues should not be set to
Patch-abandoned.


> In that case, I would expect the patch author (who should be much
> more familiar with his work than any automated script) to manually
> set it to Patch-new.  Failing that, any other developer could set
> patch-new to trigger a new test if the discussion suggests that
> the previous test results are not correct.

It's not the problem of triggering a new test, it's the problem of
making test results available to all eyes, regardless of any human
judgement.  As James uploads tests results increasingly often, and tests
run more reliably on his computer than on Grenouille, I consider giving
SFTP access on Grenouille to Patchy regular testers so they can upload
tests results, and add a hook in Patchy to do this seamlessly.


> Sure, but again I think that we can/should rely on humans manually
> saying "let's get a new set of test results for this".

Again, the problem is not getting a new set of test results, but getting
one set of test results that is available online.  What about creating a
new label category like Patchtest={todo,pass,fail}, where "todo" would
be set either when uploading a new patch or requesting a new tests run,
"pass" and "fail" would be set only if a regtest-comparison is put
online (and Grenouille would not set "fail" until we have fixed the
cause of its repeated crashes), which could be set with little or no
human intervention?  This would require no change in current Patch-*
labels semantics, which Patchy would have less to interpret.

Best,
John