Current RFC process and project decisions

php.internals

Jakub Zelenka

3 years ago
Hi, First of all the intention of this email is to try to clarify and later better define the current RFC process and project decision processes. The ultimate goal is to have a single document that would define project bylaws and except other things be a guide for new contributors to quickly get an idea how the project decisions are done. Currently the RFC voting in defined in [1]. This defines all necessary details about a discussion, proposing changes and voting. However it doesn't say anything about the actual proposals so it can be more or less anything impacting PHP. Also it doesn't say anything about implications of the accepting of the idea (e.g. if it needs be merged) or when it should be used. There is another accepted RFC about release process [2] that defines better for what RFC should be used. Specifically it states that it should be used for the features in the following sentence:
> New features or additions to the core should go through the RFC process.
It is important to note that the RFC specifically states "SHOULD" and not "MUST" which is something that has been probably applied to many smaller features that didn't go through RFC over the years. It is not exactly clear if this is actually allowed as there is a conflicting note regarding release cycle that kind of says that the feature can get to the next release as long as it gets accepted:
> The shorter release cycle also ensures that a given feature can get into
the next release, as long as the RFC has been accepted. This could be taken that only RFC accepted feature can go to the next release which kind of conflict with the "SHOULD" above. As you can see it's pretty vague and not exactly something that has been really accepted in practice. The fact is that there is also no definition of the feature so if we took it to the extreme, it could mean any technical change would also need an RFC. I think we all agree that would be really inconvenient so that's something that we should probably better clarify. We then have rules for RMs but they are mainly based around releases. I think it would be good to define that part as well but will leave it aside for now as it is much clearer and less impacting the core development. The only important thing to note is that RM decides about bugs going to the release. It seems that in the current definition, the RFC is just for blocking rejected features to get to the next release rather than requiring accepted features to be in the next release. It means if the RFC is accepted, it can get to the next release but it does not mean that it must get to the next release. Obviously if RFC is rejected, it means that it cannot get to the next release. The implication of that is that to fully accept the feature, it requires someone to merge it and no one to revert it... It means it requires some sort of consensus between core developers or at least no objections from some core developers. This is largely undefined as almost always the core developers would follow the decision in RFC but technically there is nothing in our process that would require them to do so. In terms of conflicts between developers, the RFC does not even have much meaning because of the above. It could be used just as some sort of poll and honour the decision as such but it has no deciding factor in it. This is at least how I understand the linked RFCs so please comment if you think that the analysis is incorrect or I missed anything. I think it would be good to define this somewhere so we can potentially make tweaks into it in the future. The idea is to maybe create a single document in the wiki collecting both linked RFCs with extended clarifications and maybe mirror it in git so the future proposal are easier to do. Thoughts? [1] https://wiki.php.net/rfc/voting [2] https://wiki.php.net/rfc/releaseprocess Cheers Jakub

Dan Ackroyd

3 years ago
On Mon, 1 May 2023 at 12:21, Jakub Zelenka <bukka@php.net> wrote:
> It seems that in the current definition, the RFC is just for blocking > rejected features to get to the next release rather than requiring accepted > features to be in the next release....It means it requires some > sort of consensus between core developers or at least no objections from > some core developers.
It might be better to think of the RFC process for new features to be a decision on "should this feature be in PHP?" rather than a "must this feature be in PHP?". As far as I can remember, there is a single RFC that wasn't merged in the next planned release: https://wiki.php.net/rfc/null_coalesce_equal_operator The vote was passed on 2016/04/02 so 7.1 was the next release version, but it wasn't merged until 2019/01/22 and so was in PHP 7.4. That was due to a technical problem with the implementation - I don't recall/understand the exact details. But it wasn't (afaik) a controversial decision to not merge it until the patch was good enough. This was possible to happen because we don't require a complete patch that is ready to merge. Ironing out all of the minor details for a new feature is a huge amount of work (that people in userland are usually completely oblivious to) which would increase the burden of passing RFCs. Seeing as one of the big long term problems PHP has, is that many contributors stop contributing after they get burnt out by doing an RFC, I would oppose making RFCs take more work. Once an RFC is passed the conversation changes from whether the RFC is a good idea or not, to being a conversation about how to best implement it, and whether the patch has some show-stopper issues or not.
> This is largely undefined as almost always the core > developers would follow the decision in RFC but technically there is > nothing in our process that would require them to do so.
Technically correct. Though the situation where all of the core contributors refuse to merge a PR, probably means the PR shouldn't be merged.
> The idea is to maybe create a single document in the wiki collecting both > linked RFCs with extended clarifications and maybe mirror it in git so the > future proposal are easier to do.
At the risk of being a senior programmer; what problem are you trying to solve? I mean, clearly there have been problems that could do with clarifying: * when are and what types of code cleanups allowed to happen? What notice should be given, to avoid causing work for other developers, who might have large existant branches they've been working on? * are pull-requests that merely add comments allowed? This is something some long-time core contributors have strong feelings about, which I and other people disagree strongly with, but we haven't had a framework to discuss it without shouting. And there are probably other non-userland affecting code maintenance tasks. But as far as as I'm aware, there hasn't been a problem with an RFC passing and the core contributors refusing to accept it. So please can we discuss the exact problem you want to solve, so that we can agree it's the right problem to solve, before suggesting solutions? If nothing else, some problems are unsolvable by lots of documentation and bye-laws. cheers Dan Ack

Jakub Zelenka

3 years ago
Hi, On Mon, 1 May 2023, 18:00 Dan Ackroyd, <Danack@basereality.com> wrote:
> On Mon, 1 May 2023 at 12:21, Jakub Zelenka <bukka@php.net> wrote: > > > It seems that in the current definition, the RFC is just for blocking > > rejected features to get to the next release rather than requiring > accepted > > features to be in the next release....It means it requires some > > sort of consensus between core developers or at least no objections from > > some core developers. > > It might be better to think of the RFC process for new features to be > a decision on "should this feature be in PHP?" rather than a "must > this feature be in PHP?". >
Yes this exactly what I was trying say. One of the reasons for this thread was actually to prevent misconception that accepted RFC is a final decision which I have feeling some voters might think. This is essentially mainly meant for technical changes where such RFC decision does not make any sense IMHO. Basically we decide about anything technical in RFC that explicitly allows leaving decision of merging for later.
> As far as I can remember, there is a single RFC that wasn't merged in > the next planned release: > https://wiki.php.net/rfc/null_coalesce_equal_operator > > The vote was passed on 2016/04/02 so 7.1 was the next release version, > but it wasn't merged until 2019/01/22 and so was in PHP 7.4. > > That was due to a technical problem with the implementation - I don't > recall/understand the exact details. But it wasn't (afaik) a > controversial decision to not merge it until the patch was good > enough. >
Just to clarify I don't have any problem with that. Personally I think the RFC process works fine for user facing changes and proposing changes with minimal implementation is fine in my eyes.
> This was possible to happen because we don't require a complete patch > that is ready to merge. Ironing out all of the minor details for a new > feature is a huge amount of work (that people in userland are usually > completely oblivious to) which would increase the burden of passing > RFCs. > > Seeing as one of the big long term problems PHP has, is that many > contributors stop contributing after they get burnt out by doing an > RFC, I would oppose making RFCs take more work. >
I agree and I would personally prefer not to do RFC for changes that are not controversial. Certainly I wouldn't want to make it worse.
> Once an RFC is passed the conversation changes from whether the RFC is > a good idea or not, to being a conversation about how to best > implement it, and whether the patch has some show-stopper issues or > not. > > > This is largely undefined as almost always the core > > developers would follow the decision in RFC but technically there is > > nothing in our process that would require them to do so. > > Technically correct. Though the situation where all of the core > contributors refuse to merge a PR, probably means the PR shouldn't be > merged. > > > The idea is to maybe create a single document in the wiki collecting both > > linked RFCs with extended clarifications and maybe mirror it in git so > the > > future proposal are easier to do. > > At the risk of being a senior programmer; what problem are you trying to > solve? >
The proposed single document is not meant to solve problems but improve things and make the rules clear and easy to find for new contributors. It should also remove all conflicting statements and it should also make the modification easier.
> I mean, clearly there have been problems that could do with clarifying: > > * when are and what types of code cleanups allowed to happen? What > notice should be given, to avoid causing work for other developers, > who might have large existant branches they've been working on? >
Yes this is exactly something that should not be addresed by the RFC IMO
> > * are pull-requests that merely add comments allowed? This is > something some long-time core contributors have strong feelings about, > which I and other people disagree strongly with, but we haven't had a > framework to discuss it without shouting. >
This is in some way the same as the above. Basically for me the main issue is that the RFC is the only place where conflicting technical issues between developers are decided. I think having such framework is really something that would be great to have. The TC proposal is such an attempt actually.
> And there are probably other non-userland affecting code maintenance tasks. > > But as far as as I'm aware, there hasn't been a problem with an RFC > passing and the core contributors refusing to accept it. So please can > we discuss the exact problem you want to solve, so that we can agree > it's the right problem to solve, before suggesting solutions? >
As I said it was mainly about improvements and making things clear at this stage which should allow further changes in the future. In terms of those chnges that I would like to see is to mainly having some framework for dealing with technical changes. Cheers Jakub