[RFC][Poll] Asymmetric Visibility and Accessor syntax View as Markdown

In case some people didn't see it in the previous thread, I am bumping
this poll:

https://docs.google.com/forms/d/e/1FAIpQLSefq15VvGNIXSnQaMTl3RW451w0E8oesny8c4PLqmKl8HhQ-Q/viewform

This is an informal poll on behalf of Ilija and I to determine the
syntax pattern we end up using for both Asymmetric Visibility and
Property accessor hooks. We in particular want to get feedback from
RFC voters, because as we all know "discussion on the list" is
frequently highly unrepresentative of the actual voter sentiment, and
finding out "I like the idea just not that particular syntax" after a
vote fails, well, sucks for everyone.

If you are planning to vote at all on either RFC, please take a few
moments to read the question being asked and give us feedback so we can
produce something acceptable to the broadest number of people.

Hi folks. I've now closed the previous poll regarding asymmetric visibility and have some results.

Total responses: 33
Voter responses: 11

Question 1

This question asked people to scale-rate between Swift-style (1) and C# style (7). The final counts were:

1: 6
2: 7
3: 4
4: 4
5: 3
6: 3
7: 6

Which roughly translates as "opinions are very split and all over the map, but there is a notable lean toward Swift-style." In particular, there were 13 "Strongly or very strongly Swift" responses to 9 "Strongly or very strongly C#" responses, and the middle responses also leaned slightly Swift-style.

Looking at the text responses people gave (thanks!), there were a sizable number of supporters of either model that said it "just felt better" or "feels more natural" for subjective reasons. Much of the C# support came from the fact that more PHP developers seem to also have C# experience than Swift experience, so there was simply a familiarity question at play. (Which is relevant, to be clear.)

Among the non-feel comments, most were critical of the added user-facing complexity of C# style in PHP (because of references), and therefore favored Swift.

If we restrict the results to just people with voting access, we get an average rating of 3.909, or almost a perfect tie. (Figures.)

Based on the above, Ilija and I plan to continue on with Swift-style syntax. Although it is not universally preferred, it does seem to have somewhat stronger support and more cleanly separates the two features syntactically, as they are orthogonal to each other.

Question 2

The next question was what syntax to use for the set-visibility, which was conducted with an RCV model. I computed the results using a slightly modified version of Derick's counter that is used for RFC votes. (I just tweaked it to take in data from the Google Form rather than the Wiki's output format. The counting engine is unchanged.)

I was really surprised at the results. In the first round, private(set) had a plurality lead over public(set: private) in second place:

Candidate 'private(set)': 12 → 1 2 4 8 12 13 22 25 27 28 29 30
Candidate '(set: private)': 9 → 3 10 17 19 20 21 24 26 31
Candidate 'public private': 5 → 7 9 14 16 23
Candidate 'private:set': 3 → 0 5 6
Candidate 'public:private': 3 → 11 15 18

I really expected private:set to do better, given how many people seemed to be anti-parentheses, yet the top two responses were the ones with parentheses. From examining the votes, public(set: private) was very much a "love it or hate it" response. The rest of the rounds confirmed the same results:

Round 2:
Candidate 'private(set)': 12 → 1 2 4 8 12 13 22 25 27 28 29 30
Candidate '(set: private)': 9 → 3 10 17 19 20 21 24 26 31
Candidate 'public private': 6 → 7 9 14 16 23 15
Candidate 'private:set': 5 → 0 5 6 11 18

Round 3:
Candidate 'private(set)': 13 → 1 2 4 8 12 13 22 25 27 28 29 30 6
Candidate '(set: private)': 11 → 3 10 17 19 20 21 24 26 31 5 18
Candidate 'public private': 8 → 7 9 14 16 23 15 0 11

Round 4:
Candidate 'private(set)': 20 → 1 2 4 8 12 13 22 25 27 28 29 30 6 7 9 14 16 15 0 11
Candidate '(set: private)': 12 → 3 10 17 19 20 21 24 26 31 5 18 23

And so the consensus result is indeed private(set), much to my surprise. Running the count on just voters gave the same end result.

Looking at the comments, there were again a plethora of entirely subjective responses supporting, well, everything. Several people did note that the explicit options (that had the word set in them somewhere) were better, because explicitness, and because it allows them to be abbreviated and/or used in an arbitrary order.

We are therefore going to proceed with public private(set), the same syntax as Swift and the current RFC text use.

Thank you everyone for your participation. The results clearly show that not everyone is going to be happy with the outcome no matter what, but I hope that going through this poll and sharing this level of transparency ensures that at least most people feel heard and that their viewpoints were considered.

We'll be back with an updated RFC as previously discussed as soon as Ilija is able to make it all work.

--Larry Garfield