Everything filled in

0012-unicode-strings.md

@@ -53,24 +53,50 @@ This change in definition can cause previously verified code to no longer verify
 since expressions such as `someIntValue as char` or `someCharValue + 'a'` are no longer allowed to result in
 surrogate code points.
 
+This change can also result in behavioral changes that may not be caught by verification,
+as they may not intersect with any explicit specifications.
+Here is an example of a program that will behave differently depending on the value of `/unicodeChar`,
+and hence differently on Dafny 4.0 compared to earlier versions:
+
+```dafny
+method Main() {
+ var s := "Unicode is just so \ud83d\ude0e";
+ print s, "\n" // "Unicode is just so 😎" (not affected by /unicodeChar)
+ print |s|, "\n" // /unicodeChar:0 => "21"
+ // /unicodeChar:1 => "20"
+ print s[19], "\n" // /unicodeChar:0 => "�"
+ // /unicodeChar:1 => "😎"
+}
+```
+
 A related but largely orthogonal issue has also been addressed.
 Dafny source code files must be encoded in UTF-8,
-but in previous versions of Dafny string literals could only contain printable and white-space ASCII characters,
-due to a limitation of the parser generator the toolchain uses.
+but in previous versions of Dafny,
+string and character literals could only contain printable and white-space ASCII characters,
+due to a limitation of the Coco/R parser generator the toolchain uses.
 This has been fixed, and both standard form and verbatim form string literals now allow any Unicode characters.
-A second form of escape sequence, `\UXXXXXXXX`, is now provided to support characters outside of the Basic Multilingual Plane
+A second form of escape sequence accepting a hexadecimal number with up to six digits, `\u{XXXXXX}`, 
+is now provided to support characters outside of the Basic Multilingual Plane
 using their direct Unicode code points instead of using surrogates.
 This change is fully backwards-compatible and not controlled by the `unicodeChar` flag.
 
+```dafny
+// Several different ways to express the same string literal
+var s1 := "Unicode is just so \ud83d\ude0e";
+var s2 := "Unicode is just so \u{1F60E}";
+var s3 := "Unicode is just so 😎";
+var s4 := @"Unicode is just so 😎"; // Escape sequences not supported in verbatim strings
+```
+
 The exact representation of strings at runtime, including the particular encoding,
 is an implementation detail of a particular backend, and will often be optimized for the idioms and support
 of the target environment. Enabling Unicode characters will change the target language types used to
 represent characters and strings, and hence may cause compilation errors when using additional external
 target language code.
 
 Note also that although the Unicode scalar value concept is more general than UTF-16 code units,
-it still does not always correspond to what humans will perceive as a single atomic characters when rendered;
-see the concept of grapheme clusters [here](https://unicode.org/reports/tr29/) for more information.
+it still does not always correspond to what humans will perceive as single atomic characters when rendered;
+see the concept of grapheme clusters [here](https://unicode.org/reports/tr29/) for more details.
 The proposed change to the `char` type is only intended to allow the Dafny language to safely abstract
 away from encodings, especially to support verifiably-correct code that must compile to multiple target languages.
 Providing more of the concepts defined by the Unicode standard is left out of scope for this proposal,
@@ -81,7 +107,7 @@ under the assumption that it will enable such implementations via Dafny source c
 
 ## Verification
 
-The change to the definition of `char` in the `DafnyPrelude.bpl` file is very minimal.
+The change to the definition of `char` in the `DafnyPrelude.bpl` file is minimal.
 The only change is to replace expressions such as `0 <= n && n < 65536` with a separate predicate defined as follows:
 
 ```boogie
@@ -92,9 +118,9 @@ function char#IsUnicodeScalarValue(n: int): bool {
 ```
 
 As described above in the user-facing description of this feature, this will cause verification failures by design.
-Unlike many other programming languages that have had to add Unicode support over a major version bump,
+Unlike many other programming languages that have added Unicode support over a major version bump,
 Dafny is unusually well-positioned to catch most regressions statically.
-Given this, it is not worth the effort to build any additional migration features to help customers adopt this change.
+Given this, it does not seem worth the effort to build any additional migration features to help customers adopt this change.
 
 ## Compilation/Runtime
 
@@ -190,7 +216,7 @@ The major downside is that built-in sequence operations such as `s[i]` and `s +
 effort to achieve verification, or would have to be abandoned entirely in favour of the helper methods
 provided by the shared library.
 
-## Change the definition of the string type
+## Change the definition of the `string` type
 
 The current definition of `string` as an alias for `seq<char>` is problematic because surrogate code points
 are not truly independent values in the sequence.
@@ -216,7 +242,7 @@ around encoded bytes that includes a ghost `string` value defining the actual ab
 This could make the [existing implementations of UTF-8 and UTF-16 encodings](https://github.com/dafny-lang/libraries/tree/master/src/Unicode)
 more efficient and pleasant to use.
 
-## Add a new, distinct "rune" type
+## Add a new, distinct `rune` type
 
 We could maintain the current definition of the `char` type, and introduce a new `rune` type to represent Unicode scalar values
 instead ("rune" being the term both Go and C# use for this).
@@ -238,31 +264,104 @@ that supports verification easily but is inefficient at runtime.
 The proposed lazy decoding approach should provide a good balance between the clean expression of concepts
 and efficient, unsurprising runtime behavior.
 
+## Change `char` to mean an extended grapheme cluster
+
+Rather than just augmenting `char` to represent the more abstract Unicode scalar value concept,
+we could push further and define it to mean an [*extended grapheme cluster*](https://unicode.org/reports/tr29/#Grapheme_Cluster_Boundaries),
+which is the Unicode concept closest to what humans perceive visually as a distinct, atomic "character".
+This would make operations on strings as sequences of values correspond more closely to meaningful concepts on text,
+such as `|s|` actually providing a "character count" that aligns with expectations in a word processor.
+
+Although there are definite benefits to this approach for logic that heavily processes text
+(and is the approach that the [Swift](#swift) language takes, for example),
+the definition of how scalar values are grouped into grapheme clusters
+is much more complicated and locale-specific than the simple mathematical rules of common encodings
+like UTF-8 and UTF-16. This means a higher cost and risk of bugs in adding these concepts
+to the core definition of Dafny, and a much higher likelihood of having to change
+the language definition or implementation in future versions of Unicode.
+It would also mean `char` can no longer be represented as a single integer value,
+instead needing multiple scalar values in some cases,
+which would be a large a deep change to the encoding of `char` in the Boogie prelude and translator.
+
 # Prior art
 [prior-art]: #prior-art
 
 Unicode has a long and messy history, and the approach to supporting Unicode varies dramatically across different programming languages.
-Here is the current state of several popular programming languages, especially those that Dafny currently or will likely
-soon support compiling to:
+Here is the current state of several popular programming languages, especially those that Dafny currently supports compiling to,
+or could in the future:
 
-***TODO***
+## C#:
 
-* C#:
-* Java:
-* Go:
-* JavaScript:
-* C/C++:
-* Python:
-* Ruby:
-* Rust:
-* Swift:
+`char` is an alias for the `System.Char` value type, which represents a single character as a UTF-16 code unit. 
+`string` is an alias for the `System.String` class, which represents an immutable sequence of `char` values.
+The `System.Text.Rune` struct is provided to represent any Unicode scalar value,
+and its API guarantees that invalid values (e.g. surrogates) will be rejected on construction.
+The method `String.EnumerateRunes()` produces the sequence of runes in a string via an `IEnumerator<Rune>`.
 
-# Unresolved questions
-[unresolved-questions]: #unresolved-questions
+## Java:
+
+`char` is one of the eight primitive types in Java, and also represents a UTF-16 code unit.
+In recent versions of the Java Runtime Environment, the `java.lang.String` class supports
+encoding its data either in UTF-16 or in Latin-1, where the latter is an optimization for space
+when all characters in the string are supported by this encoding. 
+
+Java does not included a dedicated type for Unicode scalar values
+and instead uses the 32-bit wide `int` primitive type.
+
+## Go:
+
+In Go a string is a read-only slice of bytes, which generally contains UTF-8 encoded bytes
+but may contain arbitrary bytes. The `rune` type is an alias for `int32`.
 
-***TODO***
+## JavaScript:
 
-# Future possibilities
-[future-possibilities]: #future-possibilities
+The JavaScript `String` type represents an immutable sequence of UTF-16 code units as integer values.
+There is no distinct type for representing individual characters.
+
+## C++:
+
+The `char` type represents bytes, and the `std::string` class from the standard library
+operates on bytes as character, and generally does not produce correct results if used
+together with any encoding other than single-byte encodings such as ASCII.
+C++11 added two new character types, `char16_t` and `char32_t`, 
+and two new corresponding `std::u16string` and `std::u32string` collection classes.
+It also provides three new kinds of string literals,
+`u8"..."`, `u"..."`, and `U"..."`,
+which produce binary values encoded with UTF-8, UTF-16, and UTF-32 respectively.
+
+## Python:
+
+In Python 2, the `str` type was used for both binary and text data.
+A separate `unicode` type was introduced to store textual data,
+and a separate mode for unicode string literals (`u"..."`).
+Python 3 changed `str` to be this `unicode` datatype instead.
+Like JavaScript, Python has no dedicated character type,
+and just uses the unbounded `int` type for Unicode scalar values.
+
+## Rust:
+
+The `char` type directly represents any Unicode scalar value, and rejects out-of-range values.
+The "string slice" type `str` is the most primitive string type,
+which is guaranteed to refers to valid UTF-8 data.
+It primarily acts as a sequence of bytes, but directly provides many methods that work with `char` values:
+`chars()` provides an iterator of `char` values,
+`get(<slice>)` returns a subslice but produces `None` if the bounds do not align with UTF-8 sequence boundaries,
+etc.
+
+## Swift:
+
+Swift defines `String` as a collection of characters,
+and unusually defines `Character` as an [*extended grapheme cluster*](https://unicode.org/reports/tr29/#Grapheme_Cluster_Boundaries),
+which is the closest concept to an atomic characeter as perceived visually by users.
+Unicode scalar values are represented with a distinct `Unicode.Scalar` structure instead.
+`String` provides `utf8`, `utf16`, and `unicodeScalars` views
+that present the contents as sequences of UTF-8, UTF-16, or UTF-32 code units respectively.
+
+# Unresolved questions
+[unresolved-questions]: #unresolved-questions
 
-***TODO***
+Is there anything more we can do to make migration easier and safer for users?
+Chance are very good that all Dafny code in existence to date either will not change behavior
+across this change, or will slightly improve because of the improved handling of surrogate code points.
+I have been unable to think of anything that would provide more value than the verifier will already provide,
+but I am open to suggestion as always!