[PATCH 1 of 4 RFC] hook: have a generic hook for transaction opening

[Matt Mackall]

On Wed, 2015-03-11 at 09:17 -0700, Ryan McElroy wrote:
> On 3/10/2015 9:15 AM, Pierre-Yves David wrote:
> >
> >
> > On 03/10/2015 06:04 AM, Augie Fackler wrote:
> >> On Tue, Mar 10, 2015 at 01:14:28AM -0700, Pierre-Yves David wrote:
> >>> This is RFC because I'm unhappy with the current names, but they are 
> >>> going
> >>> to be bikesheded anyway. So I would like to get the topic moving and 
> >>> code is
> >>> a good way to get that.
> >>>
> >>> The name in this series are:
> >>>
> >>> - txnopen (can abort)
> >>> - txnclosing (can abort)
> >>> - txnclosed
> >>
> >> I have no particular recommendations for better names (sorry), but I'm
> >> not sure what the merit of txnopen is given that we can't provide much
> >> information. A sample use case would help there, I guess.
> >
> > I've no sample usecase for txnopen, but it is very cheap to add and I 
> > guess some extension//baroque setup will be very happy to have some 
> > official way to control//record transaction opening.
> >
> > We could also move it a bit further to right after the transaction 
> > have been openened (or add a new hooks for that). So that its easy to 
> > add data in the transaction after it is open.
> >
> > Maybe we should have a hook for aborted transaction too. But yet again 
> > I've no specific usecase in mind.
> >
> >> As for the others, we could probably rework some existing
> >> functionality to be examples. verify-after-push could be done as a
> >> txnclosing, and txnclosed could be used by the notify extension, so I
> >> see merit in both of those.
> >
> > txnclosing and txnclosed are -must-have-. None of the current hook 
> > allow to catch a full transaction run with all its data. This is 
> > extremely annoying for validation or synchronization.
> >
> The only idea I have on naming is "trx" instead of "txn", which is what 
> I remember using back in my MySQL days. This is purely bikeshedding, and 
> I feel bad about it, but you did ask... sooo...

We've already got pretxncommit and pretxnchangegroup, so we're stuck
with the 'txn' and 'pretxn' bits.

I think pretxnclose and txnclose are the most consistent with our
current nomenclature.

-- 
Mathematics is the supreme nostalgia of our time.