Dear Swift-Evolution community,
A few of us have been preparing a proposal to refine the definitions of
identifiers & operators. This includes some changes to the permitted
The latest (perhaps final?) draft is available here:
We'd welcome your initial thoughts, and will probably submit a PR soon to
the swift-evolution repo for a formal review. Full text follows below.
—Jacob Bandes-Storch, Xiaodi Wu, Erica Sadun, Jonathan Shapiro
Refining Identifier and Operator Symbology
- Proposal: SE-NNNN
- Authors: Jacob Bandes-Storch <https://github.com/jtbandes>, Erica Sadun
<https://github.com/erica>, Xiaodi Wu <https://github.com/xwu>, Jonathan
- Review Manager: TBD
- Status: Awaiting review
This proposal seeks to refine and rationalize Swift's identifier and
operator symbology. Specifically, this proposal:
- adopts the Unicode recommendation for identifier characters, with some
- restricts the legal operator set to the current ASCII operator
- changes where dots may appear in operators; and
- disallows Emoji from identifiers and operators.
- Proposal: Normalize Unicode identifiers
- Unicode identifiers & operators
precursor to this document)
- Lexical matters: identifiers and operators
- Proposal: Allow Single Dollar Sign as Valid Identifier
- Free the '$' Symbol!
- Request to add middle dot (U+00B7) as operator character?
Chris Lattner has written:
…our current operator space (particularly the unicode segments covered) is
not super well considered. It would be great for someone to take a more
systematic pass over them to rationalize things.
We need a token to be unambiguously an operator or identifier - we can have
different rules for the leading and subsequent characters though.
…any proposal that breaks:
let = "moof"
will not be tolerated.
By supporting custom Unicode operators and identifiers, Swift attempts to
accomodate programmers and programming styles from many languages and
cultures. It deserves a well-thought-out specification of which characters
are valid. However, Swift's current identifier and operator character sets
do not conform to any Unicode standards, nor have they been rationalized in
the language or compiler documentation.
Identifiers, which serve as *names* for various entities, are linguistic in
nature and must permit a variety of characters to properly serve
non–English-speaking coders. This issue has been considered by the
communities of many programming languages already, and the Unicode
Consortium has published recommendations on how to choose identifier
character sets — Swift should make an effort to conform to these
Operators, on the other hand, should be rare and carefully chosen, because
they suffer from low discoverability and difficult readability. They are by
nature *symbols*, not names. This places a cognitive cost on users with
respect to both recall ("What is the operator that applies the behavior I
need?") and recognition ("What does the operator in this code do?").
every* nontrivial program defines many new identifiers, most programs do
not define new operators.
As operators become more esoteric or customized, the cognitive cost rises.
Recognizing a function name like formUnion(with:) is simpler for many
programmers than recalling what the ∪ operator does. Swift's current
operator character set includes many characters that aren't traditional and
recognizable operators — this encourages problematic and frivolous uses in
an otherwise safe language.
Today, there are many discrepancies and edge cases motivating these changes:
- · is an identifier, while • is an operator.
- The Greek question mark ; is a valid identifier.
- Braille patterns ⠟ seem letter-like, but are operator characters.
- ️:small_airplane:🂡 are identifiers, while ️:v:️:arrow_up_small:️:spades:️ are operators.
- Some *non-combining* diacritics ´ ¨ ꓻ are valid in identifiers.
- Some completely non-linguistic characters, such as ۞ and ༒, are valid
- Some symbols such as ⚄ and ♄ are operators, despite not really being
- A small handful of characters 〡〢〣〤〥〦〧〨〩 〪 〫 〬 〭 〮 〯 are valid in
- Some non-printing characters such as 2064 INVISIBLE PLUS and 200B ZERO
WIDTH SPACE are valid identifiers.
- Currency symbols are split across operators (¢ £ ¤ ¥) and identifiers
($ ₪ € ₱ ₹ ฿ ...).
This matter should be considered in a near timeframe (Swift 3.1 or 4) as it
is both fundamental to Swift and will produce source-breaking changes.
in other languages
Haskell distinguishes identifiers/operators by their general category
<http://www.fileformat.info/info/unicode/category/index.htm> such as "any
Unicode lowercase letter", "any Unicode symbol or punctuation", and so
forth. Identifiers can start with any lowercase letter or _, and may
contain any letter/digit/'/_. This includes letters like δ and Я, and
digits like ٢.
- Haskell Syntax Reference
- Haskell Lexer
Scala similarly allows letters, numbers, $, and _ in identifiers,
distinguishing by general categories Ll, Lu, Lt, Lo, and Nl. Operator
characters include mathematical and other symbols (Sm and So) in addition
to other ASCII symbol characters.
- Scala Lexical Syntax
ECMAScript 2015 ("ES6") uses ID_Start and ID_Continue, as well as
Other_ID_Start / Other_ID_Continue, for identifiers.
- ECMAScript Specification: Names and Keywords
Python 3 uses XID_Start and XID_Continue.
- The Python Language Reference: Identifiers and Keywords
- PEP 3131: Supporting Non-ASCII Identifiers
For identifiers, adopt the recommendations made in UAX #31 Identifier and
Pattern Syntax <http://unicode.org/reports/tr31/>, deriving the sets of
valid characters from ID_Start and ID_Continue. Normalize identifiers using
Normalization Form C (NFC).
(For operators, no such recommendation currently exists, although active
work is in progress to update UAX #31 to address "operator identifiers".)
Restrict operators to those ASCII characters which are currently operators.
All other operator characters are removed from the language.
Allow dots in operators in any location, but only in runs of two or more.
(Overall, this proposal is aggressive in its removal of problematic
characters. We are not attempting to prevent the addition or re-addition of
characters in the future, but by paring the set down now, we require any
future changes to pass the high bar of the Swift Evolution process.)
Swift identifier characters will conform to UAX #31
<http://unicode.org/reports/tr31/#Conformance> as follows:
UAX31-C1. <http://unicode.org/reports/tr31/#C1> The conformance
described herein refers to the Unicode 9.0.0 version of UAX #31 (dated
2016-05-31 and retrieved 2016-10-09).
UAX31-C2. <http://unicode.org/reports/tr31/#C2> Swift shall observe the
UAX31-R1. <http://unicode.org/reports/tr31/#R1> Swift shall augment
the definition of "Default Identifiers" with the following profiles:
ID_Start and ID_Continue shall be used for Start and Continue
(replacing XID_Start and XID_Continue). This excludes characters
in Other_ID_Start and Other_ID_Continue.
_ 005F LOW LINE shall additionally be allowed as a Start character.
The emoji characters 1F436 DOG FACE and 1F42E COW FACE shall
be allowed as Start and Continue characters.
(UAX31-R1a. <http://unicode.org/reports/tr31/#R1a>) The
join-control characters ZWJ and ZWNJ are strictly limited to
cases A1, A2, and B described in UAX #31. (This requirement
is covered in
the Normalize Unicode Identifiers proposal
UAX31-R4. <http://unicode.org/reports/tr31/#R4> Swift shall consider
two identifiers equivalent when they have the same normalized form under
NFC <http://unicode.org/reports/tr15/>. (This requirement is covered
in the Normalize Unicode Identifiers proposal
in the removal of some 5,500 valid code points from the identifier
characters, as well as hundreds of thousands of unassigned code points.
(Though it does not appear on this unicode.org utility, which currently
supports only Unicode 8 data, the · 00B7 MIDDLE DOT is no longer an
identifier character.) Adopting ID_Start and ID_Continue does not add any
new identifier characters.
identifier-head → [:ID_Start:]
identifier-head → _
identifier-character → identifier-head
identifier-character → [:ID_Continue:]
Swift operator characters will be limited to only the following ASCII
! % & * + - . / < = > ? ^ | ~
The current restrictions on reserved tokens and operators will remain: =, ->
, //, /*, */, ., ?, prefix <, prefix &, postfix >, and postfix ! are
The current requirements for dots in operator names are:
If an operator doesn’t begin with a dot, it can’t contain a dot elsewhere.
This proposal changes the rule to:
Dots may only appear in operators in runs of two or more.
Under the revised rule, ..< and ... are allowed, but <.< is not. We
the .. operator, permitting the compiler to use .. for a "method cascade"
syntax in the future, as supported by Dart
Motivations for incorporating the two-dot rule are:
It helps avoid future lexical complications arising from lone .s.
It's a conservative approach, erring towards overly restrictive.
Dropping the rule in future (thereby allowing single dots) may be possible.
It doesn't require special cases for existing infix dot operators in the
standard library, ... (closed range) and ..< (half-open range). It also
leaves the door open for the standard library to add analogous half-open
and fully-open range operators <.. and <..<.
If we fail to adopt this rule now, then future backward-compatibility
requirements will preclude the introduction of some potentially useful
operator → operator-head operator-characters[opt]
operator-head → ! % & * + - / < = > ? ^ | ~
operator-head → operator-dot operator-dots
operator-character → operator-head
operator-characters → operator-character operator-character[opt]
operator-dot → .
operator-dots → operator-dot operator-dots[opt]
If adopted, this proposal eliminates emoji from Swift identifiers and
operators. Despite their novelty and utility, emoji characters introduce
significant challenges to the language:
Their categorization into identifiers and operators is not semantically
motivated, and is fraught with discrepancies.
Emoji characters are not displayed consistently and uniformly across
different systems and fonts. Including all Unicode emoji
characters that don't render as emoji on Apple platforms without a variant
selector, but which also wouldn't normally be used as identifier characters
(e.g. ⏏ ).
Some emoji nearly overlap with existing operator syntax: ️:question:️
Full emoji support necessitates handling a variety of use cases for
joining characters and variant selectors, which would not otherwise be
useful in most cases. It would be hard to avoid permitting sequences of
characters which aren't valid emoji, or being overly restrictive and not
properly supporting emoji introduced in future versions of Unicode.
As an exception, in homage to Swift's origins, we permit and in
This change is source-breaking in cases where developers have incorporated
emoji or custom non-ASCII operators, or identifiers with characters which
have been disallowed. This is unlikely to be a significant breakage for the
majority of serious Swift code.
Code using the middle dot · in identifiers may be slightly more common. · is
now disallowed entirely.
Diagnostics for invalid characters are already produced today. We can
improve them easily if needed.
Maintaining source compatibility for Swift 3 should be easy: just keep the
old parsing & identifier lookup code.
on ABI stability
This proposal does not affect the ABI format itself, although the Normalize
Unicode Identifiers proposal
<https://github.com/apple/swift-evolution/pull/531> affects the ABI of
The standard library will not be affected; it uses ASCII symbols with no
on API resilience
This proposal doesn't affect API resilience.
Define operator characters using Unicode categories such as Sm and So
This approach would include many "non-operator-like" characters and doesn't
seem to provide a significant benefit aside from a simpler definition.
Hand-pick a set of "operator-like" characters to include. The proposal
authors tried this painstaking approach, and came up with a relatively
agreeable set of about 650 code points
<http://unicode.org/cldr/utility/list-unicodeset.jsp?a=[!\%24%\%26*%2B\-%2F<%3D>%3F\^|~ \u00AC \u00B1 \u00B7 \u00D7 \u00F7 \u2208-\u220D \u220F-\u2211 \u22C0-\u22C3 \u2212-\u221D \u2238 \u223A \u2240 \u228C-\u228E \u2293-\u22A3 \u22BA-\u22BD \u22C4-\u22C7 \u22C9-\u22CC \u22D2-\u22D3 \u2223-\u222A \u2236-\u2237 \u2239 \u223B-\u223E \u2241-\u228B \u228F-\u2292 \u22A6-\u22B9 \u22C8 \u22CD \u22D0-\u22D1 \u22D4-\u22FF \u22CE-\u22CF \u2A00-\u2AFF \u27C2 \u27C3 \u27C4 \u27C7 \u27C8 \u27C9 \u27CA \u27CE-\u27D7 \u27DA-\u27DF \u27E0-\u27E5 \u29B5-\u29C3 \u29C4-\u29C9 \u29CA-\u29D0 \u29D1-\u29D7 \u29DF \u29E1 \u29E2 \u29E3-\u29E6 \u29FA \u29FB \u2308-\u230B \u2336-\u237A \u2395]>
this set would require further refinement), but ultimately felt the
motivation for including non-ASCII operators is much lower than for
identifiers, and the harm to readers/writers of programs outweighs their
Use Normalization Form KC (NFKC) instead of NFC. The decision to use NFC
comes from Normalize Unicode Identifiers proposal
<https://github.com/apple/swift-evolution/pull/531>. Also, UAX #31
Generally if the programming language has case-sensitive identifiers,
then Normalization Form C is appropriate; whereas, if the programming
language has case-insensitive identifiers, then Normalization Form KC is
NFKC may also produce surprising results; for example, "ſ" and "s" are
equivalent under NFKC.
Continue to allow single .s in operators, and perhaps even expand the
original rule to allow them anywhere (even if the operator does not begin
This would allow a wider variety of custom operators (for some
interesting possibilities, see the operators in Haskell's Lens
<https://github.com/ekmett/lens/wiki/Operators> package). However, there
are a handful of potential complications:
Combining prefix or postfix operators with member access: foo*.bar would
need to be parsed as foo *. barrather than (foo*).bar. Parentheses
could be required to disambiguate.
Combining infix operators with contextual members: foo*.bar would
need to be parsed as foo *. bar rather than foo * (.bar). Whitespace
or parentheses could be required to disambiguate.
Hypothetically, if operators were accessible as members such as
MyNumber.+, allowing operators with single .s would require escaping
operator names (perhaps with backticks, such as MyNumber.`+`).
This would also require operators of the form [!?]*\. (for example . ?.
!. !!.) to be reserved, to prevent users from defining custom operators
that conflict with member access and optional chaining.
We believe that requiring dots to appear in groups of at least two,
while in some ways more restrictive, will prevent a significant amount of
future pain, and does not require special-case considerations such as the
While not within the scope of this proposal, the following considerations
may provide useful context for the proposed changes. We encourage the
community to pick up these topics when the time is right.
Re-expand operators to allow some non-ASCII characters. There is work in
progress to update UAX #31 with definitions for "operator identifiers" —
when this work is completed, it would be worth considering for Swift.
Introduce a syntax for method cascades. The Dart language supports method
whereby multiple methods can be called on an object within one expression:
foo..bar()..baz() effectively performs foo.bar(); foo.baz(). This syntax
can also be used with assignments and subscripts. Such a feature might be
very useful in Swift; this proposal reserves the .. operator so that it
may be added in the future.
Introduce "mixfix" operator declarations. Mixfix operators are based on
pattern matching, and would allow more than two operands. For example, the
ternary operator ? : can be defined as a mixfix operator with three
"holes": _ ? _ : _. Subscripts might be subsumed by mixfix declarations
such as _ [ _ ]. Some holes could be made @autoclosure, and there might
even be holes whose argument is represented as an AST, rather than a value
or thunk, supporting advanced metaprogramming (for instance, F#'s code
Diminish or remove the lexical distinction between operators and
identifiers. If precedence and fixity applied to traditional identifiers
as well as operators, it would be possible to incorporate ASCII equivalents
for standard operators (e.g. and for &&, to allow A and B). If
additionally combined with mixfix operator support, this might enable
powerful DSLs (for instance, C#'s LINQ