On Tue, Sep 17, 2019 at 3:32 PM Larry Garfield <la...@garfieldtech.com> wrote:
> Simple question for those that keep arguing that the RFC process is only > applicable to a certain subset of issues: > > OK, so what's the alternative? > > If we wanted to make a structural or governance change to PHP, what is the > process? > If we really did feel there was a reason to make a fundamental change to > the language (whatever that means), what is the process? > If we wanted to change the RFC process, what is the process? > If we don't have those, and want to set them up, what is the process for > defining the process? For the first and last one (which are kind of the same) - the answer is simply the (informal) process we had before the RFC process was enacted. That effectively meant consensus based decision making. Since we have a lot more people today, we can and probably should reuse the voting mechanism, and a pass would have to look along the lines of this: https://web.archive.org/web/20120527111218/https://wiki.php.net/rfc/voting/vote or https://wiki.php.net/rfc/abolish-short-votes If you look at all the 'Process and Policy' RFCs we've voted on, other than a couple that are miscategorized technical RFCs - they virtually all cleared a 15 to 1 bar, most of them well above that. When changing the rules - or extending the scope of the RFC process to handle things it never has before, this is what it takes. We haven't implemented any rules that bind everyone without that level of widespread agreement to this date. Consensus based decisions would work for the 3rd one as well and would probably be the simplest to enforce. It may be that for RFCs that place new limits on it (like the recent Abolish votes) a 2/3 bar would suffice - although I think it's healthy for everyone that the ratio that was reached was more along the lines of 20 to 1 than 2 to 1, in terms of everyone accepting the validity of the policy change (including the fingerful who voted against). But since determining whether a policy RFC falls in that category or not can in itself be challenging, having a single, clear high bar for introducing both changes to the Voting RFC, as well new policy rules, would probably be the simplest and probably healthiest outcome. Regarding the 2nd (fundamental / high impact changes) - the solution here too would be consensus based decision making. That's the bar we cleared in previous major changes - the deprecation of register_globals, magic_quotes and safe_mode. Now, I think Nikita does have a point that defining what constitutes a 'high impact' break vs. one that isn't very easy - especially in a formal manner. So it may make sense to have a single bar for all compatibility breaking changes, instead of a separate one for high impact ones and low impact ones. The solution might be to simply gauge the level of caring through the number of voters who took the time to vote. For instance, a change proposal that garnered 10 votes, 7 to 3, is probably not a high-impact one and it may be reasonable to accept it even if it only cleared a 2 to 1 ratio. A change proposal that garners 50 votes and is 35 in favor and 15 against (exactly the same ratio, but with a lot more voters) - is most probably a high impact one, and should clear a much higher consensus-level bar. In the meantime, formality aside, it's easy enough to 'know it when you see it'. I don't think anybody contends that changing our undefined variable behavior or deprecating short tags are high-impact breaks - in terms of the lines of code in the combined universal PHP code base that would have to be changed as a result. Other than the higher bar - I think such proposals should be required (or at the very least encouraged) to do a better impact analysis regardless. They should be tested on a set of apps (one that will attempt to represent the PHP codebase at large, not just the cutting-edge framework development), and the results should be available as a part of the RFC. Even if we can't formally compute from that data whether it constitutes high-impact or not, having that data as a part of the RFC will likely help voters determine their opinion on it - first at the level of whether they care or not, and secondly - whether they're in favor or not. This will, in turn, effect voter turnout - and help determine whether this is indeed a major change or not. In addition, I don't think we should be grouping any deprecations together into a single vote - unless that's absolutely required from a technical standpoint (i.e. doing one without the other would lead to an inconsistency). With the recent engine errors reclassification RFC, initially - the deprecation of default values for uninitialized variables wasn't even viewed as a very big deal and was grouped with the rest. It's true that this quickly became apparent and Nikita separated it after a couple of days - but I think that should be a requirement, and not up to the RFC author. I also agree with Christian - the fact that this deprecation was by far the biggest one - basically distracted everyone (myself included) from discussing the smaller ones. This means that while there are probably some issues with some of the other, smaller changes - the fact they're lumped together with others which are harmless, and the fact there was practically no discussion over any of them - means it's all too easy to vote in favor of changing the entire group. Combined with no impact analysis being available for each proposal - it's very likely that there's 'herd mentality' happening there. Putting each in a separate vote would have likely not thoroughly solved this, but it would have probably been a good first step, allowing more granular choice. I think that this particular change (requiring separate votes for each change) can be done relatively easily within our existing framework - similar to the Abolish RFCs, if there's widespread agreement. In the context of Ben's email from a few weeks ago, I'll defer to someone else to propose it if they think it makes sense. Zeev
