RFC: generator expressions

Makefile

@@ -191,6 +191,7 @@ CORE_SRCS := $(addprefix $(JULIAHOME)/, \
 		base/dict.jl \
 		base/error.jl \
 		base/essentials.jl \
+		base/generator.jl \
 		base/expr.jl \
 		base/functors.jl \
 		base/hashing.jl \

NEWS.md

@@ -4,6 +4,9 @@ Julia v0.5.0 Release Notes
 New language features
 ---------------------
 
+  * Generator expressions, e.g. `f(i) for i in 1:n` (#4470). This returns an iterator
+    that computes the specified values on demand.
+
   * Macro expander functions are now generic, so macros can have multiple definitions
     (e.g. for different numbers of arguments, or optional arguments) ([#8846], [#9627]).
     However note that the argument types refer to the syntax tree representation, and not

base/abstractarray.jl

@@ -1226,7 +1226,7 @@ function map!{F}(f::F, dest::AbstractArray, A::AbstractArray)
     return dest
 end
 
-function map_to!{T,F}(f::F, offs, st, dest::AbstractArray{T}, A::AbstractArray)
+function map_to!{T,F}(f::F, offs, st, dest::AbstractArray{T}, A)
     # map to dest array, checking the type of each result. if a result does not
     # match, widen the result type and re-dispatch.
     i = offs

base/boot.jl

@@ -341,6 +341,7 @@ unsafe_convert{T}(::Type{T}, x::T) = x
 (::Type{Array{T}}){T}(m::Int, n::Int, o::Int) = Array{T,3}(m, n, o)
 
 # TODO: possibly turn these into deprecations
+Array{T,N}(::Type{T}, d::NTuple{N,Int}) = Array{T}(d)
 Array{T}(::Type{T}, d::Int...) = Array{T}(d)
 Array{T}(::Type{T}, m::Int)               = Array{T,1}(m)
 Array{T}(::Type{T}, m::Int,n::Int)        = Array{T,2}(m,n)

base/coreimg.jl

@@ -22,6 +22,7 @@ macro doc(str, def) Expr(:escape, def) end
 
 ## Load essential files and libraries
 include("essentials.jl")
+include("generator.jl")
 include("reflection.jl")
 include("options.jl")
 

base/generator.jl

@@ -0,0 +1,21 @@
+"""
+    Generator(f, iter)
+
+Given a function `f` and an iterator `iter`, construct an iterator that yields
+the values of `f` applied to the elements of `iter`.
+The syntax `f(x) for x in iter` is syntax for constructing an instance of this
+type.
+"""
+immutable Generator{I,F}
+    f::F
+    iter::I
+end
+
+start(g::Generator) = start(g.iter)
+done(g::Generator, s) = done(g.iter, s)
+function next(g::Generator, s)
+    v, s2 = next(g.iter, s)
+    g.f(v), s2
+end
+
+collect(g::Generator) = map(g.f, g.iter)

base/iterator.jl

@@ -292,3 +292,52 @@ eltype{I1,I2}(::Type{Prod{I1,I2}}) = tuple_type_cons(eltype(I1), eltype(I2))
     x = prod_next(p, st)
     ((x[1][1],x[1][2]...), x[2])
 end
+
+_size(p::Prod2) = (length(p.a), length(p.b))
+_size(p::Prod) = (length(p.a), _size(p.b)...)
+
+"""
+    IteratorND(iter, dims)
+
+Given an iterator `iter` and dimensions tuple `dims`, return an iterator that
+yields the same values as `iter`, but with the specified multi-dimensional shape.
+For example, this determines the shape of the array returned when `collect` is
+applied to this iterator.
+"""
+immutable IteratorND{I,N}
+    iter::I
+    dims::NTuple{N,Int}
+
+    function (::Type{IteratorND}){I,N}(iter::I, shape::NTuple{N,Integer})
+        li = length(iter)
+        if li != prod(shape)
+            throw(DimensionMismatch("dimensions $shape must be consistent with iterator length $li"))
+        end
+        new{I,N}(iter, shape)
+    end
+    (::Type{IteratorND}){I<:AbstractProdIterator}(p::I) = IteratorND(p, _size(p))
+end
+
+start(i::IteratorND) = start(i.iter)
+done(i::IteratorND, s) = done(i.iter, s)
+next(i::IteratorND, s) = next(i.iter, s)
+
+size(i::IteratorND) = i.dims
+length(i::IteratorND) = prod(size(i))
+ndims{I,N}(::IteratorND{I,N}) = N
+
+eltype{I}(::IteratorND{I}) = eltype(I)
+
+collect(i::IteratorND) = copy!(Array(eltype(i),size(i)), i)
+
+function collect{I<:IteratorND}(g::Generator{I})
+    sz = size(g.iter)
+    if length(g.iter) == 0
+        return Array(Union{}, sz)
+    end
+    st = start(g)
+    first, st = next(g, st)
+    dest = Array(typeof(first), sz)
+    dest[1] = first
+    return map_to!(g.f, 2, st, dest, g.iter)
+end

base/sysimg.jl

@@ -30,6 +30,7 @@ end
 include("essentials.jl")
 include("docs/bootstrap.jl")
 include("base.jl")
+include("generator.jl")
 include("reflection.jl")
 include("options.jl")
 

doc/manual/arrays.rst

@@ -213,6 +213,40 @@ that the result is of type ``Float64`` by writing::
 
     Float64[ 0.25*x[i-1] + 0.5*x[i] + 0.25*x[i+1] for i=2:length(x)-1 ]
 
+.. _man-generator-expressions:
+
+Generator Expressions
+---------------------
+
+Comprehensions can also be written without the enclosing square brackets, producing
+an object known as a generator. This object can be iterated to produce values on
+demand, instead of allocating an array and storing them in advance
+(see :ref:`_man-interfaces-iteration`).
+For example, the following expression sums a series without allocating memory::
+
+.. doctest::
+
+    julia> sum(1/n^2 for n=1:1000)
+    1.6439345666815615
+
+When writing a generator expression with multiple dimensions inside an argument
+list, parentheses are needed to separate the generator from subsequent arguments::
+
+    julia> map(tuple, 1/(i+j) for i=1:2, j=1:2, [1:4;])
+    ERROR: syntax: invalid iteration specification
+
+All comma-separated expressions after ``for`` are interpreted as ranges. Adding
+parentheses lets us add a third argument to ``map``::
+
+.. doctest::
+
+    julia> map(tuple, (1/(i+j) for i=1:2, j=1:2), [1:4;])
+    4-element Array{Any,1}:
+     (0.5,1)
+     (0.333333,2)
+     (0.333333,3)
+     (0.25,4)
+
 .. _man-array-indexing:
 
 Indexing

src/julia-parser.scm

@@ -1401,21 +1401,22 @@
   (parse-comma-separated s parse-eq*))
 
 ;; as above, but allows both "i=r" and "i in r"
+(define (parse-iteration-spec s)
+  (let ((r (parse-eq* s)))
+    (cond ((and (pair? r) (eq? (car r) '=))  r)
+          ((eq? r ':)  r)
+          ((and (length= r 4) (eq? (car r) 'comparison)
+                (or (eq? (caddr r) 'in) (eq? (caddr r) '∈)))
+           `(= ,(cadr r) ,(cadddr r)))
+          (else
+           (error "invalid iteration specification")))))
+
 (define (parse-comma-separated-iters s)
   (let loop ((ranges '()))
-    (let ((r (parse-eq* s)))
-      (let ((r (cond ((and (pair? r) (eq? (car r) '=))
-                      r)
-                     ((eq? r ':)
-                      r)
-                     ((and (length= r 4) (eq? (car r) 'comparison)
-                           (or (eq? (caddr r) 'in) (eq? (caddr r) '∈)))
-                      `(= ,(cadr r) ,(cadddr r)))
-                     (else
-                      (error "invalid iteration specification")))))
-        (case (peek-token s)
-          ((#\,)  (take-token s) (loop (cons r ranges)))
-          (else   (reverse! (cons r ranges))))))))
+    (let ((r (parse-iteration-spec s)))
+      (case (peek-token s)
+        ((#\,)  (take-token s) (loop (cons r ranges)))
+        (else   (reverse! (cons r ranges)))))))
 
 (define (parse-space-separated-exprs s)
   (with-space-sensitive
@@ -1471,6 +1472,9 @@
                        (loop (cons nxt lst)))
                       ((eqv? c #\;)     (loop (cons nxt lst)))
                       ((eqv? c closer)  (loop (cons nxt lst)))
+                      ((eq? c 'for)
+                       (take-token s)
+                       (loop (cons (parse-generator s nxt) lst)))
                       ;; newline character isn't detectable here
                       #;((eqv? c #\newline)
                       (error "unexpected line break in argument list"))
@@ -1515,7 +1519,7 @@
 (define (parse-comprehension s first closer)
   (let ((r (parse-comma-separated-iters s)))
     (if (not (eqv? (require-token s) closer))
-        (error (string "expected " closer))
+        (error (string "expected \"" closer "\""))
         (take-token s))
     `(comprehension ,first ,@r)))
 
@@ -1525,6 +1529,9 @@
         `(dict_comprehension ,@(cdr c))
         (error "invalid dict comprehension"))))
 
+(define (parse-generator s first)
+  `(generator ,first ,@(parse-comma-separated-iters s)))
+
 (define (parse-matrix s first closer gotnewline)
   (define (fix head v) (cons head (reverse v)))
   (define (update-outer v outer)
@@ -1953,6 +1960,13 @@
                             `(tuple ,ex)
                             ;; value in parentheses (x)
                             ex))
+                       ((eq? t 'for)
+                        (take-token s)
+                        (let ((gen (parse-generator s ex)))
+                          (if (eqv? (require-token s) #\) )
+                              (take-token s)
+                              (error "expected \")\""))
+                          gen))
                        (else
                         ;; tuple (x,) (x,y) (x...) etc.
                         (if (eqv? t #\, )

src/julia-syntax.scm

@@ -1926,6 +1926,23 @@
               (lower-ccall name RT (cdr argtypes) args))))
          e))
 
+   'generator
+   (lambda (e)
+     (let ((expr   (cadr e))
+           (vars   (map cadr  (cddr e)))
+           (ranges (map caddr (cddr e))))
+       (let* ((argname (if (and (length= vars 1) (symbol? (car vars)))
+                           (car vars)
+                           (gensy)))
+              (splat (if (eq? argname (car vars))
+                         '()
+                         `((= (tuple ,@vars) ,argname)))))
+         (expand-forms
+          `(call (top Generator) (-> ,argname (block ,@splat ,expr))
+                 ,(if (length= ranges 1)
+                      (car ranges)
+                      `(call (top IteratorND) (call (top product) ,@ranges))))))))
+
    'comprehension
    (lambda (e)
      (expand-forms (lower-comprehension #f       (cadr e) (cddr e))))

test/functional.jl

@@ -178,3 +178,29 @@ let
     foreach((args...)->push!(a,args), [2,4,6], [10,20,30])
     @test a == [(2,10),(4,20),(6,30)]
 end
+
+# generators (#4470, #14848)
+
+@test sum(i/2 for i=1:2) == 1.5
+@test collect(2i for i=2:5) == [4,6,8,10]
+@test collect((i+10j for i=1:2,j=3:4)) == [31 41; 32 42]
+@test collect((i+10j for i=1:2,j=3:4,k=1:1)) == reshape([31 41; 32 42], (2,2,1))
+
+let I = Base.IteratorND(1:27,(3,3,3))
+    @test collect(I) == reshape(1:27,(3,3,3))
+    @test size(I) == (3,3,3)
+    @test length(I) == 27
+    @test eltype(I) === Int
+    @test ndims(I) == 3
+end
+
+let A = collect(Base.Generator(x->2x, Real[1.5,2.5]))
+    @test A == [3,5]
+    @test isa(A,Vector{Float64})
+end
+
+let f(g) = (@test size(g.iter)==(2,3))
+    f(i+j for i=1:2, j=3:5)
+end
+
+@test_throws DimensionMismatch Base.IteratorND(1:2, (2,3))

test/parse.jl

@@ -347,3 +347,5 @@ end
 # issue #14683
 @test_throws ParseError parse("'\\A\"'")
 @test parse("'\"'") == parse("'\\\"'") == '"' == "\""[1] == '\42'
+
+@test_throws ParseError parse("f(2x for x=1:10, y")