Files
linguist/samples/Reason/Syntax.re
2016-12-22 17:03:18 -08:00

990 lines
23 KiB
ReasonML

/* Copyright (c) 2015-present, Facebook, Inc. All rights reserved. */
[@@@autoFormat let wrap = 80; let shift = 2];
Modules.run ();
Polymorphism.run ();
Variants.run ();
BasicStructures.run ();
TestUtils.printSection "General Syntax";
/* Won't work! */
/* let matchingFunc a = match a with */
/* `Thingy x => (print_string "matched thingy x"); x */
/* | `Other x => (print_string "matched other x"); x;; */
/* */
let matchingFunc a =>
switch a {
| `Thingy x =>
print_string "matched thingy x";
let zz = 10;
zz
| `Other x =>
print_string "matched other x";
x
};
type firstTwoShouldBeGroupedInParens =
(int => int) => int => int;
type allParensCanBeRemoved =
int => int => int => int;
type firstTwoShouldBeGroupedAndFirstThree =
((int => int) => int) => int;
/* Same thing now but with type constructors instead of each int */
type firstTwoShouldBeGroupedInParens =
(list int => list int) => list int => list int;
type allParensCanBeRemoved =
list int => list int => list int => list int;
type firstTwoShouldBeGroupedAndFirstThree =
((list int => list int) => list int) =>
list int;
type myRecordType = {
firstTwoShouldBeGroupedInParens:
(int => int) => int => int,
allParensCanBeRemoved:
int => int => int => int,
firstTwoShouldBeGroupedAndFirstThree:
((int => int) => int) => int
};
type firstNamedArgShouldBeGroupedInParens =
first::(int => int) => second::int => int;
type allParensCanBeRemoved =
first::int => second::int => third::int => int;
type firstTwoShouldBeGroupedAndFirstThree =
first::((int => int) => int) => int;
/* Same thing now, but with type constructors instead of int */
type firstNamedArgShouldBeGroupedInParens =
first::(list int => list int) =>
second::list int =>
list int;
type allParensCanBeRemoved =
first::list int =>
second::list int =>
third::list int =>
list int;
type firstTwoShouldBeGroupedAndFirstThree =
first::((list int => list int) => list int) =>
list int;
type firstNamedArgShouldBeGroupedInParens =
first::(int => int)? =>
second::int list? =>
int;
/* The arrow necessitates parens around the next two args. The ? isn't what
* makes the parens necessary. */
type firstNamedArgShouldBeGroupedInParensAndSecondNamedArg =
first::(int => int)? =>
second::(int => int)? =>
int;
type allParensCanBeRemoved =
first::int? =>
second::int? =>
third::int? =>
int;
type firstTwoShouldBeGroupedAndFirstThree =
first::((int => int) => int) => int;
type noParens =
one::int => int => int => two::int => int;
type noParensNeeded =
one::int => int => int => two::int => int;
type firstNamedArgNeedsParens =
one::(int => int => int) => two::int => int;
/* Now, let's try type aliasing */
/* Unless wrapped in parens, types between arrows may not be aliased, may not
* themselves be arrows. */
type parensRequiredAroundFirstArg =
(list int as 'a) => int as 'a;
type parensRequiredAroundReturnType =
(list int as 'a) => (int as 'a);
type parensRequiredAroundReturnType =
(list int as 'a) => (int as 'a) as 'b;
type noParensNeededWhenInTuple =
(list int as 'a, list int as 'b) as 'entireThing;
type myTypeDef 'a = list 'a;
type instatiatedTypeDef = myTypeDef int => int;
/* Test a type attribute for good measure */
/* We should clean up all of the attribute tagging eventually, but for now,
* let's make it super ugly to get out of the way of all the formatting/parsing
* implementations (fewer conflicts during parsing, fewer edge cases during
* printing).
*/
type something = (
int,
int [@lookAtThisAttribute]
);
type longWrappingTypeDefinitionExample =
M_RK__G.Types.instance
(TGRecognizer.tGFields unit unit)
(TGRecognizer.tGMethods unit unit);
type semiLongWrappingTypeDefinitionExample =
M_RK__Gesture.Types.instance
TGRecognizerFinal.tGFields
TGRecognizerFinal.tGMethods;
type semiLongWrappingTypeWithConstraint =
M_RK__Gesture.Types.instance
'a
TGRecognizerFinal.tGFields
TGRecognizerFinal.tGMethods
constraint 'a = (unit, unit);
type onelineConstrain = 'a constraint 'a = int;
/* This must be in trunk but not in this branch of OCaml */
/* type withNestedRecords = MyConstructor {myField: int} */
type colors =
| Red int
| Black int
| Green int;
/* Another approach is to require declared variants to wrap any record */
/* type myRecord = MyRecord {name: int}; */
/* let myValue = MyRecord {name: int}; */
/* This would force importing of the module */
/* This would also lend itself naturally to pattern matching - and avoid having
to use `.` operator at all since you normally destructure. */
type nameBlahType = {nameBlah: int};
let myRecord = {nameBlah: 20};
let myRecordName = myRecord.nameBlah;
let {nameBlah}: nameBlahType = {nameBlah: 20};
print_int nameBlah;
let {nameBlah: aliasedToThisVar}: nameBlahType = {
nameBlah: 20
};
print_int aliasedToThisVar;
let desiredFormattingForWrappedLambda:
int => int => int => nameBlahType =
/*
fun is
pre- /firstarg\
fix /-coupled--\
|-\ /-to-prefix--\ */
fun curriedArg anotherArg lastArg => {
nameBlah: 10
};
type longerInt = int;
let desiredFormattingForWrappedLambdaWrappedArrow:
longerInt =>
longerInt =>
longerInt =>
nameBlahType =
/*
fun is
pre- /firstarg\
fix /-coupled--\
|-\ /-to-prefix--\ */
fun curriedArg anotherArg lastArg => {
nameBlah: 10
};
let desiredFormattingForWrappedLambdaReturnOnNewLine
/*
fun is
pre- /firstarg\
fix /-coupled--\
|-\ /-to-prefix--\ */
curriedArg
anotherArg
lastArg => {
nameBlah: 10
};
/*
let is
pre-
fix /-function binding name---\
|-\ / is coupled to prefix \ */
let desiredFormattingForWrappedSugar
curriedArg
anotherArg
lastArg => {
nameBlah: 10
};
/*
let is
pre-
fix /-function binding name---\
|-\ / is coupled to prefix \ */
let desiredFormattingForWrappedSugarReturnOnNewLine
curriedArg
anotherArg
lastArg => {
nameBlah: 10
};
/*
let : type t1 t2. t1 * t2 list -> t1 = ...
let rec f : 't1 't2. 't1 * 't2 list -> 't1 =
fun (type t1) (type t2) -> (... : t1 * t2 list -> t1)
*/
type point = {x: int, y: int};
type point3D = {x: int, y: int, z: int};
let point2D = {x: 20, y: 30};
let point3D: point3D = {
x: 10,
y: 11,
z: 80 /* Optional Comma */
};
let printPoint (p: point) => {
print_int p.x;
print_int p.y
};
let addPoints (p1: point, p2: point) => {
x: p1.x + p2.x,
y: p1.y + p2.y
};
let res1 = printPoint point2D;
let res2 =
printPoint {x: point3D.x, y: point3D.y};
/*
When () were used to indicate sequences, the parser used seq_expr not only
for grouping sequences, but also to form standard precedences.
/------- sequence_expr ------\
let res3 = printPoint (addPoints (point2D, point3D));
Interestingly, it knew that tuples aren't sequences.
To move towards semi delimited, semi-terminated, braces-grouped sequences:
while allowing any non-sequence expression to be grouped on parens, we make
an explicit rule that allows one single non-semi ended expression to be
grouped in parens.
Actually: We will allow an arbitrary number of semi-delimited expressions to
be wrapped in parens, but the braces grouped semi delimited (sequence)
expressions must *also* be terminated with a semicolon.
This allows the parser to distinguish between
let x = {a}; /* Record {a:a} */
let x = {a;}; /* Single item sequence returning identifier {a} */
*/
let res3 =
printPoint (
addPoints (
point2D,
{x: point3D.x, y: point3D.y}
)
);
type person = {age: int, name: string};
type hiredPerson = {
age: string,
name: string,
dateHired: int
};
let o: person = {name: "bob", age: 10};
/* Parens needed? Nope! */
let o: person = {name: "bob", age: 10};
let printPerson (p: person) => {
let q: person = p;
p.name ^ p.name
};
/* let dontParseMeBro x y:int = x = y;*/
/* With this unification, anywhere eyou see `= fun` you can just ommit it */
let blah a => a; /* Done */
let blah a => a; /* Done (almost) */
let blah a b => a; /* Done */
let blah a b => a; /* Done (almost) */
/* More than one consecutive pattern must have a single case */
type blah = {blahBlah: int};
let blah a {blahBlah} => a;
let blah a {blahBlah} => a;
let module TryToExportTwice = {
let myVal = "hello";
};
/*
Unifying top level module syntax with local module syntax is probably a bad
idea at the moment because it makes it more difficult to continue to support
`let .. in` bindings. We can distinguish local modules for `let..in` that
just happen to be defined at the top level (but not exported).
let MyModule = {let myVal = 20;} in
MyModule.x
Wait, where would this ever be valid, even if we continued to support
`let..in`?
*/
let onlyDoingThisTopLevelLetToBypassTopLevelSequence = {
let x = {
print_int 1;
print_int 20 /* Missing trailing SEMI */
};
let x = {
print_int 1;
print_int 20; /* Ensure missing middle SEMI reported well */
print_int 20
};
let x = {
print_int 1;
print_int 20;
10
/* Comment in final position */
}; /* Missing final SEMI */
x + x
};
type hasA = {a: int};
let a = 10;
let returnsASequenceExpressionWithASingleIdentifier
() => a;
let thisReturnsA () => a;
let thisReturnsAAsWell () => a;
let recordVal: int = (thisReturnsARecord ()).a;
Printf.printf
"\nproof that thisReturnsARecord: %n\n"
recordVal;
Printf.printf
"\nproof that thisReturnsA: %n\n"
(thisReturnsA ());
/* Pattern matching */
let blah arg =>
switch arg {
/* Comment before Bar */
| /* Comment between bar/pattern */ Red _ => 1
/* Comment Before non-first bar */
| /* Comment betwen bar/pattern */ Black _ => 0
| Green _ => 0
};
/* Any function that pattern matches a multicase match is interpretted as a
* single arg that is then matched on. Instead of the above `blah` example:*/
let blah =
fun
| Red _ => 1
| Black _ => 0
| Green _ => 1;
/* `fun a => a` is read as "a function that maps a to a". Then the */
/* above example is read: "a function that 'either maps' Red to.. or maps .." */
/* Thc00f564e first bar is read as "either maps" */
/* Curried form is not supported:
let blah x | Red _ => 1 | Black _ => 0;
Theres no sugar rule for dropping => fun, only = fun
*/
/* let blahCurriedX x => fun /* See, nothing says we can drop the => fun */ */
/* |(Red x | Black x | Green x) => 1 /* With some effort, we can ammend the sugar rule that would */ */
/* | Black x => 0 /* Allow us to drop any => fun.. Just need to make pattern matching */ */
/* | Green x => 0; /* Support that */ */
/* */
let blahCurriedX x =>
fun
| Red x
| Black x
| Green x =>
1 /* With some effort, we can ammend the sugar rule that would */
| Black x => 0 /* Allow us to drop any => fun.. Just need to make pattern matching */
| Green x => 0; /* Support that */
let sameThingInLocal = {
let blahCurriedX x =>
fun
| Red x
| Black x
| Green x =>
1 /* With some effort, we can ammend the sugar rule that would */
| Black x => 0 /* Allow us to drop any => fun.. Just need to make pattern matching */
| Green x => 0; /* Support that */
blahCurriedX
};
/* This should be parsed/printed exactly as the previous */
let blahCurriedX x =>
fun
| Red x
| Black x
| Green x => 1
| Black x => 0
| Green x => 0;
/* Any time there are multiple match cases we require a leading BAR */
let v = Red 10;
let Black x | Red x | Green x = v; /* So this NON-function still parses */
/* This doesn't parse, however (and it doesn't in OCaml either):
let | Black x | Red x | Green x = v;
*/
print_int x;
/* Scoping: Let sequences. Familiar syntax for lexical ML style scope and
sequences. */
let res = {
let a = "a starts out as";
{
print_string a;
let a = 20;
print_int a
};
print_string a
};
let res = {
let a = "first its a string";
let a = 20;
print_int a;
print_int a;
print_int a
};
let res = {
let a = "a is always a string";
print_string a;
let b = 30;
print_int b
};
/* let result = LyList.map (fun | [] => true | _ => false) []; */
/* OTHERWISE: You cannot tell if a is the first match case falling through or
* a curried first arg */
/* let blah = fun a | patt => 0 | anotherPatt => 1; */
/* let blah a patt => 0 | anotherPatt => 1; */
/*simple pattern EQUALGREATER expr */
let blah a {blahBlah} => a;
/* match_case */
/* pattern EQUALGREATER expr */
let blah =
fun
| Red _ => 1
| Black _ => 0
| Green _ => 0;
/* Won't work! */
/* let arrowFunc = fun a b => print_string "returning aplusb from arrow"; a + b;; */
let arrowFunc a b => {
print_string "returning aplusb from arrow";
a + b
};
let add a b => {
let extra = {
print_string "adding";
0
};
let anotherExtra = 0;
extra + a + b + anotherExtra
};
print_string (string_of_int (add 4 34));
let dummy _ => 10;
dummy res1;
dummy res2;
dummy res3;
/* Some edge cases */
let myFun firstArg (Red x | Black x | Green x) =>
firstArg + x;
let matchesWithWhen a =>
switch a {
| Red x when 1 > 0 => 10
| Red _ => 10
| Black x => 10
| Green x => 10
};
let matchesWithWhen =
fun
| Red x when 1 > 0 => 10
| Red _ => 10
| Black x => 10
| Green x => 10;
let matchesOne (`Red x) => 10;
/*
Typical OCaml would make you *wrap the functions in parens*! This is because it
can't tell if a semicolon is a sequence operator. Even if we had records use
commas to separate fields,
*/
type adders = {
addTwoNumbers: int => int => int,
addThreeNumbers: int => int => int => int,
addThreeNumbersTupled: (int, int, int) => int
};
let myRecordWithFunctions = {
addTwoNumbers: fun a b => a + b,
addThreeNumbers: fun a b c => a + b + c,
addThreeNumbersTupled: fun (a, b, c) =>
a + b + c
};
let result =
myRecordWithFunctions.addThreeNumbers 10 20 30;
let result =
myRecordWithFunctions.addThreeNumbersTupled (
10,
20,
30
);
let lookTuplesRequireParens = (1, 2);
/* let thisDoesntParse = 1, 2; */
let tupleInsideAParenSequence = {
print_string "look, a tuple inside a sequence";
let x = 10;
(x, x)
};
let tupleInsideALetSequence = {
print_string "look, a tuple inside a sequence";
let x = 10;
(x, x)
};
/* We *require* that function return types be wrapped in
parenthesis. In this example, there's no ambiguity */
let makeIncrementer (delta: int) :(int => int) =>
fun a => a + delta;
/* We could even force that consistency with let bindings - it's allowed
currently but not forced.
*/
let myAnnotatedValBinding: int = 10;
/* Class functions (constructors) and methods are unified in the same way */
class classWithNoArg = {
method x = 0;
method y = 0;
};
/* This parses but doesn't type check
class myClass init => object
method x => init
method y => init
end;
*/
let myFunc (a: int) (b: int) :(int, int) => (
a,
b
);
let myFunc (a: int) (b: int) :list int => [1];
let myFunc (a: int) (b: int) :point => {
x: a,
y: b
};
let myFunc (a: int, b: int) :point => {
x: a,
y: b
};
type myThing = (int, int);
type stillARecord = {name: string, age: int};
/* Rebase latest OCaml to get the following: And fixup
`generalized_constructor_arguments` according to master. */
/* type ('a, 'b) myOtherThing = Leaf {first:'a, second: 'b} | Null; */
type branch 'a 'b = {first: 'a, second: 'b};
type myOtherThing 'a 'b =
| Leaf (branch 'a 'b)
| Null;
type yourThing = myOtherThing int int;
/* Conveniently - this parses exactly how you would intend! No *need* to wrap
in an extra [], but it doesn't hurt */
/* FIXME type lookAtThesePolyVariants = list [`Red] ; */
/* FIXME type bracketsGroupMultipleParamsAndPrecedence = list (list (list [`Red])); */
/* FIXME type youCanWrapExtraIfYouWant = (list [`Red]); */
/* FIXME type hereAreMultiplePolyVariants = list [`Red | `Black]; */
/* FIXME type hereAreMultiplePolyVariantsWithOptionalWrapping = list ([`Red | `Black]); */
/*
/* Proposal: ES6 style lambdas: */
/* Currying */
let lookES6Style = (`Red x) (`Black y) => { };
let lookES6Style (`Red x) (`Black y) => { };
/* Matching the single argument */
let lookES6Style = oneArg => match oneArg with
| `Red x => x
| `Black x => x;
/* The "trick" to currying that we already have is basically the same - we just
* have to reword it a bit:
* From:
* "Any time you see [let x = fun ...] just replace it with [let x ...]"
* To:
* "Any time you see [let x = ... => ] just replace it with [let x ... => ]"
*/
let lookES6Style oneArg => match oneArg with
| `Red x => x
| `Black x => x;
*/
/** Current OCaml Named Arguments. Any aliasing is more than just aliasing!
OCaml allows full on pattern matching of named args. */
/*
A: let named ~a ~b = aa + bb in
B: let namedAlias ~a:aa ~b:bb = aa + bb in
C: let namedAnnot ~(a:int) ~(b:int) = a + b in
D: let namedAliasAnnot ~a:(aa:int) ~b:(bb:int) = aa + bb in
E: let optional ?a ?b = 10 in
F: let optionalAlias ?a:aa ?b:bb = 10 in
G: let optionalAnnot ?(a:int option) ?(b:int option) = 10 in
H: let optionalAliasAnnot ?a:(aa:int option) ?b:(bb:int option) = 10 in
/*
Look! When a default is provided, annotation causes inferred type of argument
to not be "option" since it's automatically destructured (because we know it
will always be available one way or another.)
*/
I: let defOptional ?(a=10) ?(b=10) = 10 in
J: let defOptionalAlias ?a:(aa=10) ?b:(bb=10) = 10 in
K: let defOptionalAnnot ?(a:int=10) ?(b:int=10) = 10 in
\ \
\label_let_pattern opt_default: no longer needed in SugarML
L: let defOptionalAliasAnnot ?a:(aa:int=10) ?b:(bb:int=10) = 10 in
\ \
\let_pattern: still a useful syntactic building block in SugarML
*/
/**
* In Reason, the syntax for named args uses double semicolon, since
* the syntax for lists uses ES6 style [], freeing up the ::.
*/
let a = 10;
let b = 20;
/*A*/
let named a::a b::b => a + b;
type named = a::int => b::int => int;
/*B*/
let namedAlias a::aa b::bb => aa + bb;
let namedAlias a::aa b::bb => aa + bb;
type namedAlias = a::int => b::int => int;
/*C*/
let namedAnnot a::(a: int) b::(b: int) => 20;
/*D*/
let namedAliasAnnot a::(aa: int) b::(bb: int) => 20;
/*E*/
let myOptional a::a=? b::b=? () => 10;
type named = a::int? => b::int? => unit => int;
/*F*/
let optionalAlias a::aa=? b::bb=? () => 10;
/*G*/
let optionalAnnot a::(a: int)=? b::(b: int)=? () => 10;
/*H*/
let optionalAliasAnnot
a::(aa: int)=?
b::(bb: int)=?
() => 10;
/*I: */
let defOptional a::a=10 b::b=10 () => 10;
type named = a::int? => b::int? => unit => int;
/*J*/
let defOptionalAlias a::aa=10 b::bb=10 () => 10;
/*K*/
let defOptionalAnnot
a::(a: int)=10
b::(b: int)=10
() => 10;
/*L*/
let defOptionalAliasAnnot
a::(aa: int)=10
b::(bb: int)=10
() => 10;
/*M: Invoking them - Punned */
let resNotAnnotated = named a::a b::b;
/*N:*/
let resAnnotated: int = named a::a b::b;
/*O: Invoking them */
let resNotAnnotated = named a::a b::b;
/*P: Invoking them */
let resAnnotated: int = named a::a b::b;
/*Q: Here's why "punning" doesn't work! */
/* Is b:: punned with a final non-named arg, or is b:: supplied b as one named arg? */
let b = 20;
let resAnnotated = named a::a b::b;
/*R: Proof that there are no ambiguities with return values being annotated */
let resAnnotated: ty = named a::a b;
/*S: Explicitly passed optionals are a nice way to say "use the default value"*/
let explictlyPassed =
myOptional a::?None b::?None;
/*T: Annotating the return value of the entire function call */
let explictlyPassedAnnotated: int =
myOptional a::?None b::?None;
/*U: Explicitly passing optional with identifier expression */
let a = None;
let explictlyPassed = myOptional a::?a b::?None;
let explictlyPassedAnnotated: int =
myOptional a::?a b::?None;
let nestedLet = {
let _ = 1;
()
};
let nestedLet = {
let _ = 1;
()
};
let nestedLet = {
let _ = 1;
()
};
let nestedLet = {
let _ = 1;
2
};
/*
* Showing many combinations of type annotations and named arguments.
*/
type typeWithNestedNamedArgs =
outerOne::(
innerOne::int => innerTwo::int => int
) =>
outerTwo::int =>
int;
type typeWithNestedOptionalNamedArgs =
outerOne::
(innerOne::int => innerTwo::int => int)? =>
outerTwo::int? =>
int;
type typeWithNestedOptionalNamedArgs =
outerOne::list string? => outerTwo::int? => int;
let x =
callSomeFunction
withArg::10 andOtherArg::wrappedArg;
let res = {
(constraintedSequenceItem: string);
(dontKnowWheYoudWantToActuallyDoThis: string)
};
let res = {
(
butTheyWillBePrintedWithAppropriateSpacing: string
);
(soAsToInstillBestDevelopmentPractices: string)
};
let x = [
(eachItemInListCanBeAnnotated: int),
(typeConstraints: float),
(
tupleConstraints: int,
andNotFunctionInvocations: int
)
];
let x = [
(butWeWillPrint: int),
(themAsSpaceSeparated: float),
(toInfluenceYour: int, developmentHabbits: int)
];
let newRecord = {
...(annotatedSpreadRecord: someRec),
x: y
};
let newRecord = {
...(annotatedSpreadRecord: someRec),
blah: 0,
foo: 1
};
let newRecord = {
...(
youCanEvenCallMethodsHereAndAnnotate them: someRec
),
blah: 0,
foo: 1
};
let newRecord = {
...(
youCanEvenCallMethodsHereAndAnnotate
them named::10: someRec
),
blah: 0,
foo: 1
};
let something: thing blah = aTypeAnnotation;
let something: thing blah = thisIsANamedArg;
let something: thing blah = aTypeAnnotation;
let something: blah = thisIsANamedArg thing;
let something: blah = typeAnnotation thing;
let newRecord = {
...(
heresAFunctionWithNamedArgs argOne::i: annotatedResult
),
soAsToInstill: 0,
developmentHabbits: 1
};
[@@@thisIsAThing];
let x = 10;
/* Ensure that the parenthesis are preserved here because they are
* important:
*/
let something =
fun
| None => (
fun
| [] => "emptyList"
| [_, ..._] => "nonEmptyList"
)
| Some _ => (
fun
| [] => "emptyList"
| [_, ..._] => "nonEmptyList"
);
/* A | B = X; */
let A | B = X;
/* A | (B | C) = X; */
let A | (B | C) = X;
/* (A | B) | (C | D) = X; */
let A | B | (C | D) = X;
/* A | B | (C | D) = X; */
let A | B | (C | D) = X;
/* (A | B) | C = X; */
let A | B | C = X;
/* A | B | C = X; */
let A | B | C = X;
/** External function declaration
*
*/
external f : int => int = "foo";
let x = {contents: 0};
let unitVal = x.contents = 210;