Re: James, don't do dev/stable

[David Kastrup]

Graham Percival <address@hidden> writes:

> On Thu, Nov 10, 2011 at 08:28:12PM +0000, address@hidden wrote:
>> I've pushed to dev/staging as
>> cd229915fc873fdb6fd0125827452cb0ba0067a7.  Since it contains a merge
>> commit (in order to have the convert-ly run separate from the actual
>> change without hurting bisection), getting it into master needs some
>> attention.  Attention we need anyway for the translation merges, so
>> let's see how we fare.
>
> oh dear.  I guess this means I need to get that patch of yours
> pushed, so that patchy can do this automatically?
>
> or if it can't be done automatically at all... argh.  Could you
> just handle it?  skip the testing; we can pick up the pieces later
> if bad stuff happens.  With a patch-set of this complexity, I
> don't think that either James or I should be handling it.  if you
> want to really test that it still compiles, could you leave your
> computer doing that overnight?

This patch-set is not complex.  It has a mini-reroute in the middle in
order to leave a single uncompilable commit out of the main line.

If this fails, we get _one_ commit that may cause a false positive with
git bisect.  In short: it is the simplest case we can get, and when
things go wrong, the consequences will possible at some point of time be
a minor annoyance to somebody.

I am fine with doing the hard work when I get the message "David, I
tested dev/staging, commit 04324324xxxxxxxxxx, and it is good."

But it is a bit pointless, since instead of mailing me this, one can
just do
git push origin 04324324xxxxxxxxxx:master
and it will _do_ all the checking I want.  The one thing that can happen
is that git refuses to do this (because master is ahead of staging with
some commit), and _then_ a mail to David "Eh, staging got out of sync
with master, can you fix that?" is ok.  David fixes, Patchy can go ahead
again.

I can do the bounce and merge manually until we are sure that the tool
that James uses is back to trivial: it should just checkout dev/staging,
test it, and try pushing to master.  At some point of time it might be
nice if it also recognized when _staging_ was strictly _behind_ master,
and let it catch up, and give _master_ the checking that it never got
(so that we know when to give somebody a kick somewhere), but that is a
different task.

-- 
David Kastrup