[swift-evolution] [Pitch] Retiring `where` from for-in loops

Xiaodi Wu xiaodi.wu at gmail.com
Fri Jun 10 11:28:51 CDT 2016

On Fri, Jun 10, 2016 at 6:10 AM, Karl <razielim at gmail.com> wrote:

> -1
> * Swift is explicitly a C-family language. In most or all other C-family
> languages, for loop statements allow specification of conditions for
> exiting the loop but not for filtering. Therefore, Swift's use of `where`
> is unprecedented and needs to be learned anew by every user of Swift.
> When was this decided? I distinctly remember some bloke under Craig
> Federighi’s hair saying that it was time to “move beyond” C and essentially
> ditch legacy conventions which no longer make sense.

I think you misunderstood my argument here. I don't mean that we should
yoke ourselves to C conventions, and we should absolutely ditch C
convention when it doesn't make sense. The big-picture argument here is
that `where` doesn't pass the bar of correcting a C convention that no
longer makes sense.

FWIW, on the topic of syntax choices, here is what Chris Lattner had to say
on this list:

Kevin got it exa*c*tly right, but I’d expand that last bit a bit to:
> “… pi*c*king the one that is most familiar to programmers in the extended
> *C* *family* is a good idea.["]
> The extended *C* *family* of language (whi*c*h in*c*ludes *C*, *C*++, Obj
> *C*, but also *C*#, Java, Javas*c*ript, and more) is
> an extremely popular and widely used set of languages that have a lot of
> surfa*c*e-level similarity. I
> don’t *c*laim to know the design rationale of all of these languages, but
> I surmise that this is not an
> a*c**c*ident: programmers move around and work in different languages,
> and this allows a non-expert in the
> language to understand what is going on. While there are things about *C*
> that are really unfortunate IMO
> (e.g. the de*c*larator/de*c*laration spe*c*ifier part of the grammar)
> there is a lot of goodness in the basi
> *c*operator set, fo*c*us on dot syntax, and more.
> I do agree that there are some benefits to dit*c*hing bra*c*es and
> relying on indentation instead, but there are
> also downsides. Deviating from the *C* *family* in this respe*c*t would
> have to provide **overwhelmingly** large
> advantages for us to take su*c*h a plunge, and they simply don’t exist.

> As I understand it, Swift is a new language with new conventions. It is
> desirable to align as many of those as possible with existing conventions
> so as to be easily learned, but if you limit Swift to other languages
> conventions you deny it any identity. Did Python ask anybody’s opinion
> before dropping curly-braces? Did people learn whatever Perl is supposed to
> be? Look at C’s hieroglyphic for loops!

I don't think we disagree here.

> Realistically, “for … in … while” is not going to cause incredible
> confusion. Removing it would cause a lot of frustration. You can’t on the
> one hand say our users are comfortable with the axioms of C’s hieroglyphic
> loops, and on the other hand say “for x in y while" is confusing.
> Again, as I said, once you've mastered something, by definition you find
> it not confusing. Why should we doom x% of new users to writing a loop
> incorrectly at least once when we don't have to?
> Ah, but if you’re not “doomed” to failing once, how will you ever master
> anything? Nobody knew how to write a C for-loop until someone showed them
> (and even then…). Nobody is going to just open a REPL and start writing
> code, with zero prior understanding of what Swift syntax looks like.

The thought here is along the lines of what Chris said, quoted above, and
repeated here: "The extended C family of language [...] is an extremely
popular and widely used set[;] programmers move around and work in
different languages, and [aligning to expectations arising from other C
family languages] allows a non-expert in the language to understand what is
going on." By contrast, the `where` clause violates that expectation and I
do not see "overwhelmingly large advantages" for doing so.

> * The word "where" does not consistently imply `break` or `continue`. In
> current Swift, `where` implies `break` in the context of a `while` loop and
> `continue` in the context of a `for` loop. Some users intuitively guess the
> correct meaning in each context, while others guess the wrong meaning.
> Therefore, the only way to learn for sure what `where` means in any context
> is to read the rulebook. That, by definition, means that this is
> unintuitive.
> I didn’t even know while loops supported “where”. I can’t even imagine
> what that would look like, or how I would reason about one if I saw one. I
> Googled around a little bit and couldn’t find any examples. If they exist,
> sure, go ahead, get rid of them. Nobody will miss them.

Actually, we had a *huge* chain where there were definitely people who said
they would miss them, even though as you said it appears scarcely used and
not very well known. The pernicious problem with it was that it forced even
unrelated boolean assertions to be chained with `where`, as in:

while let x = iterator.next() where y < z { ... }

> It definitely makes sense on ‘for’, though. Lots and lots of people will
> miss that; it’s a pretty well-known feature.

(See Erica's statistics below.)

Also, after everything you said, it’s still not unintuitive. That is not
> how languages work at all. Languages spoken by human beings are always
> ambiguous to some extent, and we use context to determine which meaning is
> correct:
> (Quote from
> https://research.googleblog.com/2016/05/announcing-syntaxnet-worlds-most.html
> )
> One of the main problems that makes parsing so challenging is that human
> languages show remarkable levels of ambiguity. It is not uncommon for
> moderate length sentences - say 20 or 30 words in length - to have
> hundreds, thousands, or even tens of thousands of possible
> syntactic structures. A natural language parser must somehow search through
> all of these alternatives, and find the most plausible structure given
> the context. As a very simple example, the sentence "Alice drove down the
> street in her car" has at least two possible dependency parses:
> The first corresponds to the (correct) interpretation where Alice is
> driving in her car; the second corresponds to the (absurd, but
> possible) interpretation where the street is located in her car. The
> ambiguity arises because the preposition “in" can either
> modify drove or street; this example is an instance of what is
> called prepositional phrase attachment ambiguity.
> Even algebra is not completely unambiguous - you need to use BODMAS rules
> to disambiguate potential meanings.
> It’s this context which I think you’re missing when zooming in at the word
> “where”:
> - The context that this is a variation of a ‘for x in y’ loop. We know
> that it loops through every item in ‘y' and assigns it ‘x’. It is literally
> Section 2 of the 'Swift Tour' - you learn how to assign a variable, and
> then you learn about the “for x in y” loop. Everybody should recognise it.
> - The context that ‘x’ is the subject, so ‘where’ is clearly a condition
> for x to fulfill
> - The context that ‘where’ occurs after ‘in’, so it follows the order in
> which its written: ‘for every x in y, where such-and-such is true, do …”
> - The “for x in y” loop is a data-driven loop. It doesn’t even have a loop
> index. It is not like a C for loop and you shouldn’t expect to reason about
> it that way.
> * There are other ways to break from a loop or continue to the next
> iteration without performance penalty. Nearly all of these serve more
> general purposes than a `where` clause. Some of these (such as `if` or
> `guard`) would already be familiar to a new user before they encounter
> loops, assuming a typical order for learning a programming language. Many
> of these (such as filtering methods on collections, or simply `if`) would
> be familiar to a user of another C-family language. Therefore, the `where`
> clause provides no independent utility, is not more discoverable than its
> alternatives, and is not required for progressive disclosure of an
> important facility to a learner (i.e. a simplified syntax for those who may
> not be ready for the advanced concepts needed to use a more fully-featured
> alternative).
> You say the points in favour of removal are not handwavey, but I’m still
> not convinced. “There are other ways to go to where this shortcut goes” is
> not reasoning. And I’d definitely argue that it is more discoverable than
> the ‘guard’ statement. The guard statement is stone-dead last at the end of
> a massive “Control-Flow” page. I would guess that most first-time readers
> skip those topics for later.

You cannot say the same about `if`.

> The point here is that this is not a slippery slope. If `where` offered
> independent utility, then some confusion alone probably wouldn't be enough
> to justify removal, though it may justify some consideration for change.
> However, as the extensive discussion has shown, there is nothing `where`
> can do that something else can't do better. I know you like it for style,
> but that's not sufficient grounds for keeping something confusing, IMO.
> It’s more readable. It does that better.

Earlier in this thread and others, I gave my reasoning where I disagree
with this assertion about being more readable.

> The tests also seem to show that (bizarrely) it’s also slightly faster
> than the alternatives.

I don't believe there has been any demonstration that it's faster than
`guard` or `if`. I would be shocked if that were the case.

> Karl
> On 10 Jun 2016, at 08:25, Xiaodi Wu via swift-evolution <
> swift-evolution at swift.org> wrote:
> On Fri, Jun 10, 2016 at 12:48 AM, Brandon Knope <bknope at me.com> wrote:
>> On Jun 10, 2016, at 1:08 AM, Xiaodi Wu via swift-evolution <
>> swift-evolution at swift.org> wrote:
>> On Thu, Jun 9, 2016 at 9:45 PM, Dany St-Amant <dsa.mls at icloud.com> wrote:
>>> Le 9 juin 2016 à 14:55, Xiaodi Wu via swift-evolution <
>>> swift-evolution at swift.org> a écrit :
>>> There have been, in previous threads, several examples given where users
>>> of Swift have found the behavior of `where` to be misleading and confusing.
>>> Sorry Xiaodi, but beside you (on multiple instances), and recently
>>> Erica, I have do not recall hearing that many voices saying that 'where' is
>>> confusing.
>> Shawn Erickson wrote this to the list just yesterday:
>> "I support your position on the use of where and while/when being
>> confusing in the loop statement. I (and I know others) have for example
>> used where in a loop statement mistakenly thinking it would terminate the
>> loop early but of course learned that it basically filters what causes the
>> loop body to be executed. After the fact that made sense to me but it
>> didn't click at first."
>> Couldn't we find examples of anyone being confused at any syntax?
>> Especially with an unfamiliar construct in a new language.
>> If people find the new proposed syntax confusing, do we pull that too? At
>> what point do we stop?
> That is why I favored (1) removal of the confusing syntax altogether; and
> (2) this proposal, which involves aligning the confusing syntax with an
> existing syntax. In short, no new syntax to get confused about.
> Yes, there's was maybe even less voices stating that it is not confusing,
>>> but which group is more vocal?
>>> Maybe I have been recently corrupt by Solid SQL queries:
>>> select * from PEOPLE_TABLE where AGE_FIELD = 100
>>> Or by my (likely) broken English:
>>> The places where I had the most fun
>>> But, to me, where can only suggest some filtering (thus tag to a for ..
>>>  in .., continue if not matching).
>> I'm glad that you find it very clear. I do as well. That does not mean it
>> is clear to everyone.
>> I still have yet to see widespread confusion of this. A few people
>> learning swift here or there, but once they learn the syntax...do they
>> still find it confusing?
> I expect some concrete data on stuff like this...especially with proposed
>> syntax changes.
>> Without concrete examples, what would stop one from coming in here and
>> waving their hands around to push *what they like* through?
> Here's what's not handwavy:
> Conclusion: the `where` clause is unprecedented, unintuitive, provides no
> independent utility, is not more discoverable than alternatives, and is not
> required for pedagogical reasons; however, it has been used incorrectly by
> at least some users. Therefore, it is harmful and ought to be removed or
> reformed.
> I know there's a linguist on the list, maybe he could comment on whether
>>> or not using 'where' as a filter is proper or an abomination.
>>> I do not think that because something is confusing to some, or at first,
>>> that it warrant removal from the language.
>> It is a very bad sign if something is confusing at first, especially to a
>> significant proportion of users. It's true by definition that once you have
>> mastered something you are no longer confused by it.
>> Again, where is this significant proportion of users? I don't mean to
>> hound you on this, but I am genuinely curious where this is all coming from.
> We were talking about the hypothetical something here and what the bar
> should be for removal from the language. My response is that being
> confusing at first sight *is* a legitimate consideration for removal from
> the language. If something turns out to be a confusing way to describe a
> straightforward concept, then the more widespread the confusion, the more
> urgent its removal.
>> The burden of evidence is on the proposers of these ideas.
>> As has been stated on this list, education is a valid and important
>> consideration for Swift. If something is confusing rather than difficult
>> (and the *concept* of filtering a list is not at all a difficult concept),
>> and if the same underlying concept can already be invoked in alternative
>> and equivalent ways that are not confusing, then it's a no-brainer that the
>> confusing thing is harmful to the language and should be removed on that
>> basis alone.
>> What is clear to one person may be confusing to another. There is no
>> perfect syntax that will not make it confusing for some users.
>> ----
>> I really think it is important to come armed with more information with
>> these proposals. It's easy to say a significant proportion of people are
>> confused but it would make me much more comfortable to see this data to
>> back it up.
>> What if we are spinning our wheels for no reason on a feature that *most*
>> don't find confusing? What if we make a bigger proportion of those who did
>> understand it more confused now?
>> Brandon
>> By analogy, Chinese and Japanese share difficult writing systems. Yet
>> many people use those languages daily without difficulty. Does that mean
>> there's not a problem? Far from it: in fact, you'll find that many
>> intelligent people have devoted their life's work to mitigating the issue.
>> Both Chinese and Japanese underwent a round of simplification in the 20th
>> century. Think about it: real languages used for daily life by a
>> significant fraction of the world's population were revamped for the
>> purpose of increasing accessibility to new learners.
>> The by-value/by-reference is well define, but can be confusing at first.
>>> Same goes for eager/lazy processing, or escaping vs non-escaping closure,
>>> or even the difference between closure and function. But no one suggest to
>>> remove them.
>> Value types vs. reference types is a concept (and a moderately advanced
>> one), eager vs. lazy processing is a concept (and a moderately advanced
>> one), and closures are a concept (and definitely an advanced one).
>> Filtering a collection is a concept as well, and no one is suggesting its
>> removal. We are proposing to simplify and rationalize the syntax by which
>> filtering is invoked. If there were a way to dramatically simplify the
>> syntax surrounding value types and reference types so as to diminish
>> confusion, you can absolutely guarantee that there would be proposals to
>> change the syntax. If I could think of one tomorrow, you'd see a thread
>> tomorrow about it. I don't think I'm that smart though.
>>> Dany
>>> In fact, the first of these proposals began with a question: how does
>>> one write arbitrary Boolean assertions after a let binding? The answer (use
>>> `where`) was found to be misleading and confusing.
>>> I think you're being unfair to say that these proposals have no purpose
>>> other than an academic consistency.
>>> On Thu, Jun 9, 2016 at 13:29 Jon Shier via swift-evolution <
>>> swift-evolution at swift.org> wrote:
>>>>         As time goes on, I’m feeling more and more that these
>>>> consistency proposals are sorely misguided. Frankly, unless the syntax is
>>>> confusing or misleading, even once the developer has learned the guiding
>>>> principles of Swift, consistency is not a good argument for change. This
>>>> proposal is the perfect example of this. No one will find the use of
>>>> “where” in loops confusing, aside from those who will wonder why it was
>>>> removed from if statements. There is no misleading behavior or confusing
>>>> syntax here. This is just consistency for consistency’s sake. Once this
>>>> proposal is done, then another will be made to remove “where” from another
>>>> place in the language. Then another and another until it’s gone completely
>>>> and a very useful part of the language is removed in the name of
>>>> consistency. Which really just comes down to “where” isn’t used here, so it
>>>> can’t be used there anymore. It’s death by a thousand cuts.
>>>> Jon Shier
>>>> > On Jun 9, 2016, at 1:16 PM, Erica Sadun via swift-evolution <
>>>> swift-evolution at swift.org> wrote:
>>>> >
>>>> >
>>>> >> On Jun 9, 2016, at 11:11 AM, Charlie Monroe <
>>>> charlie at charliemonroe.net> wrote:
>>>> >> See my latest post - included results with -Ofast. But still, using
>>>> filter and lazy.filter is 10+% slower, which were the suggested
>>>> alternatives to `where`.
>>>> >>
>>>> >>
>>>> >
>>>> > I need to correct this misapprehension.
>>>> > My suggested alternative to where was and remains `guard`.
>>>> >
>>>> > -- E
>>>> >
>>>> > _______________________________________________
>>>> > swift-evolution mailing list
>>>> > swift-evolution at swift.org
>>>> > https://lists.swift.org/mailman/listinfo/swift-evolution
>>>> _______________________________________________
>>>> swift-evolution mailing list
>>>> swift-evolution at swift.org
>>>> https://lists.swift.org/mailman/listinfo/swift-evolution
>>> _______________________________________________
>>> swift-evolution mailing list
>>> swift-evolution at swift.org
>>> https://lists.swift.org/mailman/listinfo/swift-evolution
>> _______________________________________________
>> swift-evolution mailing list
>> swift-evolution at swift.org
>> https://lists.swift.org/mailman/listinfo/swift-evolution
> _______________________________________________
> swift-evolution mailing list
> swift-evolution at swift.org
> https://lists.swift.org/mailman/listinfo/swift-evolution
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20160610/120f3aa2/attachment.html>

More information about the swift-evolution mailing list