Am 16.09.2019 um 21:32 schrieb Nikita Popov [email protected]:

Thanks for chiming in Rasmus. A few brief thoughts on recent discussions:

• The RFC process encompasses language changes (breaking or non-breaking),
  as well as policy and process decisions. We have a very wide selection of
  precedent cases that affirm this.

It feels to me that it gradually changed into that.
And while the initial RFCs seemed innocuous policy and process decisions they are now used as a precedent. Should the first non-technical RFCs have been examined more closely due to this? Maybe. But just because it wasn't done doesn't mean the situation cannot be reassessed again. Just because there are now policy and process RFCs does not mean we could take a step back and limit RFCs again.

Just to be clear: I'm not demanding anything, I'm just wary of the "that's just the way it is, look at previous examples" argument.

• The "undefined variables" vote that sparked the current controversy
  currently sits at 29 in favor of exception and 20 against. That's
  significantly below the acceptance threshold for RFCs. Things are working
  as they should: The question has been discussed, put up to vote, and the
  vote has decided against this course of action (as of this writing, though
  I expect this to be representative of the final result.)

I agree but disagree at the same time:
First of all the discussion was unpleasant which I don't see as "working as it should".

And while I do think the discussion about undefined variables did clear some things up I also think it distracted from other points in the RFC.
I personally chose my battle to focus on undefined variables because it was the biggest pain point. But there are lots of other conversions to Exceptions which were left in the main bulk of things even though they have similarities with undefined variables. A foreach over null might indicate a problem but it is well defined and might occur in correctly working code.
Yes, a warning might be appropriate but stopping execution is a different beast again.

• As a personal failure, I should have made the voting option for "undef
  vars throw exception" be "undef vars throw warning in PHP 8 and exception
  in PHP 9", which would have provided for a long-term migration timeline for
  affected code. I apologize for pushing an unnecessarily aggressive option
  here.

Maybe it was quite the opposite: It forced the discussion to happen now. And while it was unpleasant I'm worried that otherwise it would have legitimised making undefined variables an Exception because "we already promoted it to a warning so we all agree that it is bad" which would be wrong.
Sure, maybe by the time PHP 9 comes around people agree that an Exception is the right thing to do. But using warnings as a door-opener for exceptions is something to be considered very carefully.

• Chris

Le 16 sept. 2019 à 21:32, Nikita Popov [email protected] a écrit :

• Discussion threads on this mailing list have been very unpleasant
  recently. I am unwilling to actively participate in them in this form. This
  is bad for everyone, but particularly for opponents of proposals. It means
  that we cannot establish the necessary discourse to explore improvements or
  alternatives. The recent propensity to suppress certain discussion topics
  entirely, as well as the use of overwhelming quantity to disproportionately
  push a position, contribute to the unproductive discussion environment.

The discussion would have been less unpleasant if everyone sought consensus, that is, tried to find a solution that is appropriate for everyone, instead of trying to convince others that their opinion is the right one.

There is a qualitative difference between consensus and unanimity. Unanimity means that a solution is preferred by everyone — which is reasonably not possible. Consensus means that a solution is acceptable to everyone, even when it is not the best one for everyone. This is mostly possible, but only if everyone tries to satisfy not only themself, but also the others. But consensus cannot be measured by vote.

For example, throwing a TypeError for uninitialised variables cannot reach consensus, because it is not appropriate for those that rely on implicitly initialised variables. Triggering an E_WARNING is nearer to a possible consensus.

I am not optimistic that there would be a mindset change in the direction of seeking consensus for every participating party of the discussion. However, I am suggesting that if a RFC try to seek a solution that is at least acceptable to everyone, there will be less frustration and less irritation from the minority.

—Claude