RFC Process: Should we hold two different votes?

Le 05/11/2014 00:58, Andrea Faulds a écrit :

Good evening,

A lot of RFCs have been rejected because, while they proposed a feature people would like, the details were problematic. This has lead to these features sometimes being considerably delayed.

So, in order to do something about this, here’s an idea: Why not hold two different votes on an RFC, similar to how legislation is passed in the UK’s House of Commons? The first is on whether the general principle of the RFC is sound. Once that’s passed, it’s clear the feature is wanted, so time can then be spent scrutinising the details of the proposal and making them acceptable. Then, a second vote can be held, which approves the RFC as a whole, and its patch (like our current votes do). This way:

• Authors know quickly whether a feature has sufficient support (reading internals doesn’t necessarily tell you anything, as votes and numbers of positive/negative emails rarely align), without having to necessarily have done everything before the final vote
• Bad ideas are rejected sooner
• Good ideas with flawed implementations may succeed in the first vote and fail in the second, meaning there’s a clear agreement that it’s wanted but not with this implementation, allowing another RFC with an improved approach to perhaps be made later

Does this sound like a good idea?

I really like this idea, which might help the RFC process in several ways:

People (especially those who are not involved much in PHP internals now)
might hesitate less before working on a "principle" RFC, knowing they
won't have to code a feature (which might take days or more) that may
just not be accepted.

People not knowing C / PHP internals might suggest more "principle"
RFCs. And if those pass with an overwhelming "yes", it might be easier
for them to find support from current maintainers to help work on the
implementation.

Also, it might be the right time to have different people voting, like:

• Community (representatives of frameworks, UGs, ...), who are close
  to users but don't know C / PHP internals, should be able to vote on the
  "principle" RFCs.
• While voting on "implementation" RFCs might be restricted to people
  who are closer to PHP -- mostly, those who can already vote today (like
  people with karma, or working on doc)

Like you said in your last point, it also allows for a more iterative
work on implementation, if a "principle" is accepted but an
"implementation" is not.

Though, I think there might not always be a need for "implementation"
RFCs: if the implementation of a "principle" is trivial enough (like,
for instance, adding a simple new function or changing the default value
of a parameter), having to vote on the "implementation" could be a too
heavy process.

--
Pascal MARTIN
http://blog.pascal-martin.fr
@pascal_martin

-----Original Message-----
From: Andrea Faulds [mailto:ajf@ajf.me]
Sent: Wednesday, November 05, 2014 1:59 AM
To: PHP internals
Subject: [PHP-DEV] RFC Process: Should we hold two different votes?

Why not hold two
different votes on an RFC, similar to how legislation is passed in the UK’s
House of Commons?

I can think of two key reasons for why not:

1. It's more hassle. Unlike parliament members, this work is voluntary and
   we should be respectful of people's time. While dividing this to two votes
   may save the RFC proposer some time, especially in case his proposal doesn't
   pass the initial vote (so that they won't have to invest efforts in creating
   a detailed proposal), for everyone else, this is going to be a lot more
   hassle.

2. In my experience, there are few cases where the details of the RFC don't
   affect my yes/no decision. Voting 'in principle' without having the details
   bares very little significance. Also, running two votes may also create
   perception in some people's minds that if an RFC passed the initial vote,
   it's now only a matter of deciding between options, when again, given that
   the devil is in the details, once these details are known there may not be a
   majority in favor of the feature in the first place.

All in all, I don't see why this extra step cannot be simply replaced with
an informal discussion on internals, before the RFC is even written, to
gauge people's response.

Zeev