[swift-evolution] [Review] SE-0168: Multi-Line String Literals

Wed Apr 12 05:42:06 CDT 2017

On 12.04.2017 13:16, Thorsten Seitz via swift-evolution wrote:
>> Am 12.04.2017 um 10:11 schrieb Adrian Zubarev via swift-evolution 
>> <swift-evolution at swift.org <mailto:swift-evolution at swift.org>>:
>>
>> Great explanation thank you Brent. I’m convinced about the closing delimiter now. =)
>>
>> -------------------------------------------------------------------------------------
>>
>> If I understood correctly what Xiaodi Wu meant in his reply, then we could simplify 
>> the whole multi-line string literal and also remove the need of disabling the 
>> stripping algorithm.
>>
>> We should ban these examples completely:
>>
>> |"""Hello·world!"""|
>>
> 
> Being able to use ""“ for single line strings containing lots of " is useful in 
> itself and explained in the motivational section of the proposal:
> "Tripled string literals can also do double duty as a syntax for handling short 
> string literals with many internal quotation marks“
> 
> -Thorsten

Yes, I also think the single line string can be very useful and we should not 
disallow it.

But I agree that we should disallow multi-line cases when we have text on the same 
line with leading or trailing """ because this complicates the mental modal and adds 
confusion points.

I.e. I suggest to allow only two forms:
1. Single line:  """this is "just" text""" (no line end will be inserted)
2. Multiline, where leading and trailing """ has no text after/before them and *all* 
the text is in lines *between* triple quotes:
"""
   first line
   second line
"""

One can use backslash at the line end to emulate all other needed cases. Like:

"""
   first line \
   second line\
"""

will produce "first line second line"

> 
>> |"""Hello↵ world!""" |
>> |"""Hello↵ world!↵ """ |
>> |"""↵ Hello↵ world!""" |
>>
>> Instead an empty multi-line string literal would look like this:
>>
>> |"""↵ """ |
>>
>> To fix the above example you’d need to write it like this:
>>
>> |"""↵ Hello·world!\↵ """ |
>> |"""↵ Hello↵ world!\↵ """ |
>>
>>   * Each line in between the delimiters would add implicit new lines if not
>>     disabled by a backslash.
>>   * The trailing precision is also handled by the backslash.
>>   * The indent is handled by the closing delimiter.
>>   * It’s easier to learn/teach.
>>   * It’s easier to read, because most of the time the line where the starting
>>     delimiter is, is filled with some other code.
>>
>> |let myString = """↵ ⇥ ⇥ Hello↵ ⇥ ⇥ world!\↵ ⇥ ⇥ """ |
>>
>> Now that would be a true multi-line string literal which needs at least two lines 
>> of code. If you’d need a single line literal,|""|is the obvious pick.
>>
>>
>>
>>
>> -- 
>> Adrian Zubarev
>> Sent with Airmail
>>
>> Am 12. April 2017 um 02:32:33, Brent Royal-Gordon (brent at architechies.com 
>> <mailto:brent at architechies.com>) schrieb:
>>
>>>
>>>> On Apr 11, 2017, at 8:08 AM, Adrian Zubarev via swift-evolution 
>>>> <swift-evolution at swift.org <mailto:swift-evolution at swift.org>> wrote:
>>>>
>>>> That’s also the example that kept me thinking for a while.
>>>>
>>>> -------------------------------------------------------------------------------------
>>>>
>>>> Overall the proposal is a great compromise to some issues I had with the first 
>>>> version. However I have a few more questions:
>>>>
>>>>   * Why can’t we make it consistent and let the compiler add a new line after the
>>>>     starting delimiter.
>>>>
>>>> | let string = """↵ Swift↵ """ // result ↵Swift↵ |
>>>>
>>>> If one would would the behavior from the proposal it’s really easy to add a 
>>>> backslash after the starting delimiter.
>>>>
>>>> | let string = """\↵ Swift\↵ """ // result Swift |
>>>>
>>>> This would be consistent and less confusing to learn.
>>>>
>>> That would mean that code like this:
>>>
>>> print("""
>>> A whole bunch of
>>> multiline text
>>> """)
>>> print("""
>>> A whole bunch more
>>> multiline text
>>> """)
>>>
>>> Will print (with - to indicate blank lines):
>>>
>>> -
>>> A whole bunch of
>>> multiline text
>>> -
>>> -
>>> A whole bunch more
>>> multiline text
>>> -
>>>
>>> This is, to a first approximation, never what you actually want the computer to do.
>>>>
>>>>   * Can’t we make the indent algorithm work like this instead?
>>>>
>>>> |let string = """\↵ ····<tag>↵ ······content text↵ ····</tag>""" // Indent starts 
>>>> with the first non space character // result <tag>↵ ··content text↵ </tag> |
>>>>
>>>> The line where the closing delimiter is trims all space chapters and the indent 
>>>> for the whole multi-line string is starting at the point where the first 
>>>> non-space chapters is in that line.
>>>>
>>> We could; I discuss that briefly in the very last section, on alternatives to the 
>>> indentation stripping we specify:
>>>
>>> • Stripping indentation to match the depth of the least indented line: Instead of 
>>> removing indentation to match the end delimiter, you remove indentation to match 
>>> the least indented line of the string itself. The issue here is that, if all lines 
>>> in a string should be indented, you can't use indentation stripping. Ruby 2.3 does 
>>> this with its heredocs, and Python's dedent function also implements this behavior.
>>>
>>> That doesn't quite capture the entire breadth of the problem with this algorithm, 
>>> though. What you'd like to do is say, "all of these lines are indented four 
>>> columns, so we should remove four columns of indentation from each line". But you 
>>> don't have columns; you have tabs and spaces, and they're incomparable because the 
>>> compiler can't know what tab stops you set. So we'd end up calculating a common 
>>> prefix of whitespace for all lines and removing that. But that means, when someone 
>>> mixes tabs and spaces accidentally, you end up stripping an amount of indentation 
>>> that is unrelated to anything visible in your code. We could perhaps emit a 
>>> warning in some suspicious circumstances (like "every line has whitespace just 
>>> past the end of indentation, but some use tabs and others use spaces"), but if we 
>>> do, we can't know which one is supposed to be correct. With the proposed design, 
>>> we know what's correct—the last line—and any deviation from it can be flagged *at 
>>> the particular line which doesn't match our expectation*.
>>>
>>> Even without the tabs and spaces issue, consider the case where you accidentally 
>>> don't indent a line far enough. With your algorithm, that's indistinguishable from 
>>> wanting the other lines to be indented more than that one, so we generate a result 
>>> you don't want and we don't (can't!) emit a warning to point out the mistake. With 
>>> the proposed algorithm, we can notice there's an error and point to the line at fault.
>>>
>>> Having the closing delimiter always be on its own line and using it to decide how 
>>> much whitespace to strip is better because it gives the compiler a firm baseline 
>>> to work from. That means it can tell you what's wrong and where, instead of doing 
>>> the dumb computer thing and computing a result that's technically correct but useless.
>>>>
>>>> PS: If we’d get this feature in Swift, it would be nice if Xcode and other IDEs 
>>>> which supports Swift could show space characters that are inside a string literal 
>>>> (not other space character <- which is already supported), so it would be easier 
>>>> to tell what’s part of the string and what is not.
>>>>
>>> That would be very nice indeed. The prototype's tokenizer simply concatenates 
>>> together and computes the string literal's contents after whitespace stripping, 
>>> but in principle, I think it could probably preserve enough information to tell 
>>> SourceKit where the indentation ends and the literal content begins. (The 
>>> prototype is John's department, though, not mine.) Xcode would then have to do 
>>> something with that information, though, and swift-evolution can't make the Xcode 
>>> team do so. But I'd love to see a faint reddish background behind tripled string 
>>> literal content or a vertical line at the indentation boundary.
>>>
>>> In the meantime, this design *does* provide an unambiguous indicator of how much 
>>> whitespace will be trimmed: however much is to the left of the closing delimiter. 
>>> You just have to imagine the line extending upwards from there. I think that's an 
>>> important thing to have.
>>>
>>> -- 
>>> Brent Royal-Gordon
>>> Architechies
>>>
>>
>> _______________________________________________
>> swift-evolution mailing list
>> swift-evolution at swift.org <mailto: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
> 