[swift-evolution] Inconsistencies related to prepositions

[Mike Sanderson]

On Wed, Aug 2, 2017 at 1:49 PM, Xiaodi Wu via swift-evolution <
swift-evolution at swift.org> wrote:

> ...
>
> This topic was reviewed extensively at the beginning of Swift Evolution.
> The idea is that type information does not need to be repeated in the
> label. Since the locale is of type Locale, “with” is preferred over
> “withLocale” because “withLocale: Locale” is redundant. As this has been
> long approved by the community, it’s not in scope to roll it back.
>

Thanks to everyone who contributed to the list to make the Swift 3 API
guidelines possible. I think the Swift 3 API Guidelines were a tremendous
accomplishment, and I appreciate the actual beauty that can result from its
sparse code.

It seems reasonable that 1 1/2 years after the proposal, more than 1 year
after beta release, and after seeing the API Guidelines applied out in the
world (or not) and applying them ourselves (or not), to bring up the
subject again and ask if it can be improved. I think Jon gave some good
examples of weaknesses that can result, and Robert gave some good
suggestions.

Changing existing code is probably out of scope to avoid language
churn--and it definitely would contradict Swift 4's goal of preserving
source stability. The Great Re-Renaming? No thanks. (Thought I think The
Great Renaming unfairly got the blame for later migration issues.)  Still,
these messages today were substantive suggestions for how to improve Swift
API naming conventions. I am re-reading the threads from Jan and Feb 2016,
which seems appropriate to inform a discussion that builds on them, but it
would be weird to regard those conversations as a reason to shut down any
mention of the subject ever again. This seems like exactly the kind of
subject that should be revisited, and could form a proposal later in the
fall.

As someone who admires the API guidelines and how they make Swift read
elegantly, I want to make them even better and easier to use. I myself
think that the English language is too ungainly to support rules for a
language as elegant as Swift. So I hope we can continue to have this
conversation.

Mike Sanderson





>
>>    1.
>>    2. `dict.remove(at:)` is a bit of a tougher case. It is mutating, so
>>    it has to be in the present tense. Yet it does return an element, so maybe
>>    it should state that? I think the deciding vote is that it’s
>>    @discardableResult, which means it should be treated as primarily removing
>>    the element; returning it is secondary. Maybe to indicate the returning of
>>    the element, it should be called `dict.pop(at:)`. “Pop" implies the element
>>    popped, and probably implies @discardableResult in most programmers’ minds
>>    as well.
>>
>>
> Brent had a draft proposal to revise the names of collection methods to
> improve the situation here. There is room for improvement.
>
>
>>    1.
>>    2. `date.timeInterval` is another tricky one. IMO the most natural
>>    choice is `date.timeIntervalSince(_:)`. I don’t think the date argument
>>    label is necessary because given a date, you can only compute a time
>>    interval since… another date. See the `array.append(contentsOf:)` example
>>    above. And although `since` is a preposition, it should be part of the
>>    method name, and not an argument label, because dates do not just “have”
>>    time intervals. A database can fetch a record, but a date does not have a
>>    time interval — it has a time interval since (another date).
>>    3. This should definitely be `forEach`. `array.each` would be another
>>    name for `array.everyElementSatisfiesCondition(_ f:
>>    (Element)->Bool))`. “For each” specifies that you will do something *for
>>    each* element as opposed to asking a question about each element. In
>>    English, “for each thing in my bucket…” will probably be followed by “do
>>    this thing”. You could say “for each thing in my bucket, it’s true that…”
>>    but you’d probably just say “everything in my bucket is…"
>>
>>
>> So, if I had to come up with my own rules for naming conventions:
>>
>>
>>    1. Methods should first state what they do, then how they do it. You
>>    fetch a record using an ID; you don’t fetch a thing using a record ID. Ask
>>    the question “does the calling variable have such-and-such a thing” or “can
>>    the calling variable do such-and-such a thing”, and be specific about
>>    exactly what the thing is. If yes, than that thing is a good candidate for
>>    a method name. A database can fetch, sure, but more specifically it can
>>    fetch records, and that’s about as specific as you can get. A date doesn’t
>>    have a collection of time intervals sitting around, but it can compute the
>>    *time interval since* another date. Strings can be capitalized (with a
>>    locale), dictionaries can remove (at an index), arrays can append (the
>>    contents of some collection).
>>    2. When the preposition/argument label makes it clear what sort of
>>    argument it takes, then it’s fine to omit the type. However if the argument
>>    label is vague, then the argument type should be included. In this case the
>>    type is not just a type specifier — it’s really part of the English
>>    description. You don’t say “append the contents of collection that bucket
>>    to my array”, but any sentence about capitalizing a string with a locale
>>    would use the word “locale” explicitly, either in the name of the variable
>>    or otherwise — “capitalize that string with your specified locale”,
>>    “capitalize that string with locale XYZ”, etc. The English language doesn’t
>>    really have the notion of implicitly locale-like objects the way it has
>>    implicitly collection-like objects such as buckets, bags, etc, so the fact
>>    that a locale is being used should be explicit while the fact that a
>>    collection is being used may be implicit.
>>    As an aside, I feel that “using Locale” would be better than “with
>>    Locale”, but that’s a conversation for another time.
>>    3. Three nouns may be required in a name, as in the fetch record
>>    example. The calling variable is always the subject, and there may be
>>    either one or two objects in play. If there are two, then one should go in
>>    the method name; the other may or may not be made explicit in the argument
>>    label, depending on clarity (see the above point). When a database fetches
>>    a record using its ID, you have only two choices: db.fetchRecord(with:) or
>>    db.fetchRecord(withID:). In this case, clarity would lead to the second
>>    choice, as you don’t know what you’re fetching the record *with*. In the
>>    date example, the calling date is the subject, and it is *producing a time
>>    interval since* another date. So again, two choices:
>>    date.timeIntervalSince(date:) or date.timeIntervalSince(_:). Since you can
>>    only find the time interval between two dates, `date:` can be omitted.
>>
>>
>> It’s hard to make hard-and-fast rules around this because language is far
>> from hard-and-fast itself. But I think these would be good guidelines.
>>
>>
>>
>> On Aug 2, 2017, at 11:44 AM, Jon Gilbert via swift-evolution <
>> swift-evolution at swift.org> wrote:
>>
>> When Swift 3 launched, it introduced a new concept of placing the
>> preposition inside the parentheses. (See discussion here:
>> https://lists.swift.org/pipermail/swift-evolution/Week
>> -of-Mon-20160208/009523.html).
>>
>> I'm fine with that, however this change got implemented in an
>> inconsistent manner, making Swift and its APIs more vague, thereby
>> decreasing code clarity.
>>
>> I was hoping to see these inconsistencies get cleared up with Swift 4,
>> however they were not. I realized that perhaps I could have spoken up and
>> taken more of an active role in Swift Evolution. So... better late than
>> never. Here are my thoughts, towards consistency and disambiguation in
>> Swift.
>>
>> Disclaimer: I’m sure the last thing anyone wants are more changes to the
>> APIs. However, I also don't think these inconsistencies should be left in
>> place forever. I have tried to go back and read through as much of the
>> relevant discussions on list as I could, and it seems like
>>
>> Please let me some give examples below. After that is some discussion and
>> a proposal.
>>
>> *EXAMPLES*
>>
>> Take for example the preposition "with." A preposition is meaningless
>> without a subject and an object. Look at the following three sentences:
>>
>>  (A) "A boy with the dog Molly walked across the street."
>>
>>  (B) "With the dog Molly walked across the street."
>>
>>  (C) "A boy with Molly walked across the street."
>>
>> Sentence (A) is longest, but it makes perfect sense.
>>
>> Sentence (B) is nonsensical and grammatically incorrect because while the
>> word "with" takes two arguments, only one is present. If there was a comma
>> after "dog," then it would make sense; "With the dog, Molly walked across
>> the street" is fine. But if we were to assume that Molly is not the dog, in
>> this case, we'd actually be wrong.
>>
>> Sentence (C), while grammatically correct, leaves it unclear whether
>> Molly is a dog, a girl, or… something else.
>>
>> The reason for this is, whenever a preposition is used in English, it
>> almost always takes a dyadic form, relating a subject to the preposition's
>> object. The two most common dyadic formats are:
>>
>> *<subject> [<preposition> <object of preposition>]*
>> <The boy> [<with> <the dog>] crossed the street.
>>
>> *[<preposition> <** object of preposition**>] <subject>*
>> [<In> <space>], <no one> can hear you scream.
>> [<On> <the Moon>] are <many craters>.
>>
>> Now, in Objective C through Swift 1 and 2, prepositions' dyadic nature
>> were generally respected in method signatures. However, Swift 3's migration
>> of the preposition inside the parentheses also seems to have been
>> accompanied by the stripping away of either the subject, the prepositional
>> object, or both—according to no discernible pattern. For example:
>>
>> (1) CloudKit:
>>
>> old: myCKDatabase.fetchRecordWithID(recordID)
>> new: myCKDatabase.fetch(withRecordID: recordID)
>> *(subject "Record" got removed)*
>>
>>
>> (2) String:
>>
>> old: myString.capitalizedStringWithLocale(_: myLocale)
>> new: myString.capitalized(with: myLocale)
>> *(subject "String" and prep. object "Locale" both got removed)*
>>
>>
>> (3) Dictionary:
>>
>> old: myDict.removeAtIndex(myIndex)
>> new: myDict.remove(at: myIndex)
>> *(subject "element" already missing from both; prep. object "Index" got
>> removed)*
>>
>>
>> (4) Date:
>>
>> old: myDate.timeIntervalSinceDate(myDate)
>> new: myDate.timeIntervalSince(date: myDate)
>> *(subject "timeInterval" and object "Date" both still present; but oddly,
>> preposition "since" got left outside of the parentheses)*
>>
>> (5) Array:
>>
>>     old: myArray.forEach({ thing in code})
>>
>> new: myArray.forEach() { thing in //code }
>>
>>             *(preposition “for” is outside of the parentheses)*
>>
>> *DISCUSSION OF EXAMPLES*
>>
>> These four changes are inconsistent with each other regarding whether the
>> subject, prepositional object, or both were removed, and we don't even have
>> consistency as to whether the preposition got moved inside the parentheses
>> (see A below).
>>
>> As well, these changes generally removed the dyadic arguments to the
>> preposition, which removes cues necessary to disambiguate the prepositional
>> relationship, decreasing code readability (see B below).
>>
>> *(A) Inconsistency*
>>
>> The inconsistency between the examples is shown in the bold text of each
>> example, but lets go over why this matters. It matters because any language
>> is easier to learn the more consistently it sticks to its own rules.
>> Autocomplete is our great savior, but still, if we were being consistent,
>> then the new method signatures would have been:
>>
>> (1) myCKDatabase.fetchRecord(withRecordID:)
>> (2) myString.stringCapitalized(withLocale:)
>> (3) myDictionary.elementRemoved(atIndex:)
>> (4) myDate.timeInterval(sinceDate:)
>> (5) myArray.each(inClosure: )
>>
>> Side note: for plain English readability, we might prefer
>> elementRemoved(fromIndex:) and stringCapitlized(accordingToLocale:).
>>
>> Although I do understand removing "string" from the latter was to reduce
>> redundancy in function/method declarations, we only make one declaration,
>> yet we make many calls. So increasing ambiguity in calls does not seem like
>> a good trade-off for decreased boilerplate in declarations. More often than
>> not it's calls that we're reading, not the declarations—unless of course
>> the call was ambiguous and we had to read the declaration to make sense out
>> of it. So perhaps we might question if increased ambiguity is an overall
>> good thing.
>>
>> Side note: example (5), .forEach, does seem like a very exceptional case,
>> and possibly a sacred cow. See the Proposal section for further discussion
>> of this.
>>
>> *(B) Increased Ambiguity*
>>
>> In all of these changes, one of or both parts of the dyadic arguments of
>> the preposition have been excised from the method signatures. This
>> increases ambiguity in Swift 3/4 vs. Objective C and Swift 2, especially in
>> closures, as I will explain below.
>>
>> In example (1), the old method argument makes grammatical sense because
>> "record with RecordID" follows the format, *<subject> [<preposition>
>> <object of preposition>]*, just like "boy with dog tag" or "cow with
>> black spots." This improves code readability, because when you read this,
>> you know that this function will give you a record matching a particular
>> recordID. However in Swift 3, we have the equivalent of, "with a
>> recordID"—i.e. is *implied* that the thing being fetched is a record.
>>
>> This isn't a problem you're reading the method signature in the header,
>> and it's not a problem when you're writing the code, because you'll get
>> warnings and errors if you try to assign to the wrong type.
>>
>> However this removal of explicit contextual cues from the method
>> signature harms readability, since now, the compiler will let people write
>> code like:
>>
>> { return $0.fetch(withRecordID:$1) }
>>
>> Clearly, the onus is now on the developer not to use cryptic, short
>> variable names or NO variable names. However, spend much time on GitHub or
>> in CocoaPods and you will see an increasing number of codebases where
>> that's exactly what they do, especially in closures.
>>
>> Another problem is that the compiler doesn't care if you write:
>>
>> { ambiguousName in
>> let myRecordID = ambiguousName.fetch(withRecordID:myID)
>> return myRecordID }
>>
>> This is highly problematic because someone reading this code will have no
>> reason to expect the type of "myRecordID" not to be CKRecordID. (In fact,
>> it's CKRecord.) There is also no way to clarify what ambiguousName is since
>> closures don't have argument labels at all... but that's another problem
>> altogether.
>>
>> Turning now to example (2), "myString.capitalized(with:myLocale)"
>> sacrifices BOTH the subject and the prepositional object. We have now lost
>> any enforced contextual cues, fully orphaning "with", allowing and even
>> encouraging a final closure argument like:
>>
>> { return $0.capitalized(with: .current)) }
>>
>> What is $0? Reading this, and being familiar with Swift, you will likely
>> assume it's a string, but will you remember that .current is a Locale? The
>> compiler knows, but why was it a good idea to strip out that information?
>>
>> We also have examples like:
>>
>> { return $0.draw(with:$1) }
>>
>> What is $0? What is $1? This is a real Apple API, BTW.
>>
>> We could also have:
>>
>> {array, key in
>> let number = array.remove(at:key)
>> return number }
>>
>> This will compile and run even though number will be a tuple key-value
>> pair, array will be a dict, and key will be an index type! This may seem
>> like a ridiculous example, but I have literally seen things like this.
>>
>> *DISCUSSION*
>>
>> Making computer code more like natural language and thus more
>> approachable to non-computer-science types was clearly one of the primary
>> goals of Apple's method names in Objective C.
>>
>> I've been a fan of the general direction in which Swift has taken things,
>> but I'm starting to see code that is quite unreadable as a result of the
>> stripping away of useful information from code, as if more white space will
>> improve our ability to make sense out of our arcane invocations.
>>
>> The point of code readability is for humans to be able to read and
>> understand code with minimal extra effort. Explicit information is
>> important because humans would like to read code without having to dig
>> through five different headers and read lots of documentation files. This
>> directly increases costs, because when a new developer comes onto a
>> project, the longer it takes them to understand the codebase, the more it
>> will cost the company.
>>
>> I don't believe the intent with these changes was to increase ambiguity;
>> at WWDC '16 we were told it was to increase clarity. I'm just not sure
>> that's the consistent effect we actually received.
>>
>> What we got instead is more times we will need to CMD-click things in
>> XCode to get a clue. That's especially annoying because you're at XCode's
>> mercy as to whether it actually wants to jump to a given declaration or
>> not. This seems pretty hit or miss in my experience, even within a basic
>> playground page, and is especially troublesome with jumping to declarations
>> within CocoaPods. Selecting some code to get documentation in the side
>> panel seems equally unreliable.
>>
>> Perhaps CMD-click problems have to do with a project setting, but is
>> there really a time when you would ever NOT want to have these
>> capabilities? Why is it tied into a setting? Further, when you're reading
>> through a repo on GitHub itself, you don't have CMD-click capability; you
>> want the code to be as clear as possible, without lots of ambiguity.
>>
>> *From Ambiguous to Cryptic*
>>
>> Orphaning method signatures by stripping useful return type and argument
>> type information wouldn't be so bad if variables were all named
>> descriptively, but that is a strangely optimistic hope for a language
>> that's as paranoid about safety that it was specifically designed to
>> prevent many categories of common mistakes.
>>
>> The increased ambiguity in Swift from the removal of prepositional
>> components has been further compounded by other forms of ambiguity that are
>> increasingly everywhere in Swift, and which developers have taken copious
>> advantage of to the point of abuse. Examples are:
>>
>> - arguments labelled as "_" reduce code clarity
>> - calls to functions that take closures as the final argument can leave
>> the final argument label out, decreasing readability
>> - closures have lack explicit type information, leaving it up to variable
>> names to let us know what's going on; however, the use $# syntax and/or
>> cryptically named variables often leaves it totally ambiguous
>> - generic types in method signatures using a single letter give no clue
>> as to the type's purpose
>> - function types can't have argument labels, which might give a sense of
>> context or informative cues, even if they added boilerplate
>> - inferred nested type parents (like the ability to do .current instead
>> of being forced to do Location.current)
>>
>> *Why is removing context and clue good?*
>>
>> Is it possible that this was an overreaction to the perceived
>> over-verbosity of Objective C? I'm honestly curious. I will try to read
>> back through the archives, but perhaps someone can give me a good summary.
>>
>> In the old Objective C style, method names could be quite long, but I
>> never felt a lack of context that made code unclear. The method signatures
>> were basically like sentences without spaces, and they seemed to follow
>> pretty reliable rules. So reading them actually made grammatical sense,
>> improving code readability. I could always tell what a call was doing.
>>
>> While I won't deny that Objective C went overboard sometimes in
>> verbosity, and while few would dispute that we should clean up its style,
>> could it be we are going too far making Swift encourage code that's
>> ambiguous to everyone except the compiler? Could it be a mistake for Swift
>> to err so far on the side of inferring everything and stripping away
>> contextual information?
>>
>> A first-time reader of some unseen code is not going to be able to infer
>> things like the original programmer or the compiler can. This increases the
>> possibility of bugs (and exploits) sneaking in "under the radar."
>>
>> *PROPOSAL*
>>
>> To add more clarity back to Swift, I'd propose the following rules be
>> applied consistently across the language:
>>
>> • Every time a preposition shows up in a function/method signature, there
>> is an explicit subject and object of the preposition, except in the
>> following exceptions.
>>
>> • The subject of a preposition can be left out if the function/method is
>> mutating AND doesn't return a value, such as Array's .append(contentsOf:)
>> method. This method would become: .append(contentsOfCollection:) so that
>> it's clear what we're doing here, even within a closure.
>>
>> • No preposition should be used when it would add no extra meaning
>> between what’s inside and outside the parentheses. For example,
>> userFont(ofSize fontSize:) should just be userFont(size: fontSize), because
>> “of” is completely superfluous here. AVFragmentedMovie’s
>> track(withTrackID:) could just as easily be track(forTrackID:) or
>> track(TrackID:) which tells us that the preposition is superfluous and so
>> why not just have it be track(id:).
>>
>> • If the preposition goes with a verb in the method signature, the verb
>> should be the last thing before the opening parenthesis (e.g. String's
>> .capitalized function would now look like name.stringCapitalized(accordingToLocale:
>> .current)... no pun intended).
>>
>> • In the case of .forEach, since it’s almost identical in function to
>> .map, perhaps a solution could be to eliminate .forEach entirely, and
>> simply say that anytime .map’s result is unused (or assigned to _), then
>> the compiler will automatically use .forEach functionality to optimize
>> performance. For example if in Swift 5 you did _ = myArray.map() {
>> print($0) }, this would be the same as doing myArray.forEach() { print($0)
>> } currently. As well, _ = could be inferred.
>>
>> Note: part of the problem is obviously that parameter names are never
>> used in function calls when there is an argument label, but often, the
>> argument label is now just a preposition, while its prepositional object is
>> the (now hidden) parameter name. Perhaps we can think of a novel solution
>> to this?
>>
>> I'm interested to know what people think about this. Perhaps there are
>> other ideas or ways to add clarity back in. Thanks for reading this long
>> message.
>>
>> Jon
>> _______________________________________________
>> 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/20170802/092e61a4/attachment.html>