No pull for mtd?

[Brian Norris]

On Wed, Jul 17, 2013 at 6:03 PM, David Woodhouse <dwmw2 at infradead.org> wrote:
> On Wed, 2013-07-17 at 16:26 -0700, Brian Norris wrote:
>> Is there any hassling the rest of us can do in the future to avoid
>> this situation?
>
> Certainly can't hurt :)

I was kinda wondering about specifics :) Like more email? More capital
letters in our emails? Track down your phone number and call you in
the middle of the night?

>> Ezequiel gave a timely reminder via email (2 or 3 days
>> left in the merge window). We can shout louder and earlier, but it's
>> hard to tell if we're shouting into an empty forest sometimes.
>
> Earlier is better. If I haven't got a tree lined up in good time
> *before* the merge window opens, Linus isn't going to want to know.
> So if the merge window is already open and I'm slacking and/or changing
> nappies, that's a bit late already.

Yeah, I realize now that 3 days is a bit late. And I didn't really
notice our missing the merge window myself until it was nearly too
late :)

> Occasionally when I've failed to get things together in time for the
> merge window, I have handled that by just looking at Artem's "l2-mtd"
> tree and submitting what's in there as-is, without making any changes.
> Or submitting a strict subset of it, up to the *important* part but
> leaving out some of the more dubious later commits which I want to
> change or reject.
>
> Normally though, I use Artem's l2-mtd tree a bit like patchwork. He
> hoovers up most things from the list and even does some preliminary
> fixes on them — so it's *much* more convenient for me than raw patchwork
> would be. And it's close *enough* to what I'm actually going to accept,
> that it's reasonable for it to be in linux-next as it is.

I guess I didn't fully grasp the relation between you and Artem. It
seemed to me like Artem's tree was practically the pull request tree,
that his status as "not the maintainer" simply gave him the luxury to
rebase more freely, and that your sign-off (or sometimes pull request
w/o adding sign-offs) was mostly just a rubber stamp. But I was
slightly off, I suppose. Anyway, I do appreciate the editorial
control, whether it's Artem or you, and I would not be hasty to
discard that.

> Artem has historically expressed a reticence to have direct commit
> access to the main MTD tree (I think he does actually *have* it, in
> fact, but he's never used it). I think he prefers that I continue to
> have the final say after he does the first pass, and I think that model
> generally works out OK as a division of labour.
>
> So... other than putting a few more hours in my day (and ringfencing
> them so they don't get used up by all the *other* things I need a few
> more hours each day for, like sleeping as I should be doing right now),
> what can we do?

Lengthen your day? Maybe a time machine? Hire a body double to go to
your "real" job?

> Brian, Huang, thanks for volunteering to help. It's much appreciated.
>
> I think the way forward, if Artem is willing, might be the following:
>
> Turn the l2-mtd tree into a group-access repository, so all of us can
> push to it (or rebase/reorder/etc). Try to keep it (as Artem already
> does to a certain extent) in order of "good" and then "staging" patches.

I did not realize that there was an order other than chronological in
which Artem kept l2-mtd. This might be a little harder to manage if
the distinction between "good" and "staging" are unclear, and there
are up to 3 people rebasing and pushing. We'll definitely need Artem's
agreement for anything of this sort.

> It's that sorting that's important; I should be able to just pull the
> 'good' ones directly into the main tree as a stable commit which *will*
> to go Linus.

My suggestion, to resolve some of the potential conflicts of too many
writing to the same branch and to make the sorting clear: perhaps a
separate branch that will hold the "good" stuff (named 'for-3.12' for
example) with minimal rebasing. Then l2-mtd's 'master' will rebase on
top of that to include "staging" fixes that are more likely to be
fixed-up, dropped, or delayed.

> And then after a while, once we're all happy that I never
> actually end up rejecting or even tweaking patches in the "good" pile,
> we'll just start using the main linux-mtd.git tree as the "good" pile
> and you can push there directly. I'll try to be more explicit about what
> I change and why, in cases where I would previously have silently fixed
> things up.
>
> I'll probably still keep doing a final pass on it for a while even after
> that, and constructing the pull request. But I'm not averse to letting
> others do that too, once we've reached the point where I'm sure it'd be
> as I want it.
>
> How does that sound?

I think that mostly sounds OK. I can't speak for everyone, but it
seems that for some of us, the process just needs to be able to adapt
to changes in job situations (where at least you and Artem aren't even
employed in MTD-related work anymore), vacations, babies, absences, or
any of the other reasons that MTD participation ebbs and flows. So
bringing others into the process should help.

I guess we'll see what Artem thinks when he's back from vacation and
what Huang thinks. But I'm game to try this out.

Thanks for sacrificing sleep for the response.

Brian