implement named tuples

Based on #16580, also much work done by quinnj.

`(a=1, ...)` syntax is implemented, and `(; ...)` syntax is
implemented but not yet enabled.

NEWS.md

@@ -11,6 +11,9 @@ New language features
     a function argument name, the argument is unpacked into local variables `x` and `y`
     as in the assignment `(x, y) = arg` ([#6614]).
 
+  * Named tuples, with the syntax `(a=1, b=2)`. These behave very similarly to tuples,
+    except components can also be accessed by name using dot syntax `t.a` ([#22194]).
+
  * Custom infix operators can now be defined by appending Unicode
    combining marks, primes, and sub/superscripts to other operators.
    For example, `+̂ₐ″` is parsed as an infix operator with the same

base/boot.jl

@@ -120,7 +120,7 @@ export
     # key types
     Any, DataType, Vararg, ANY, NTuple,
     Tuple, Type, UnionAll, TypeName, TypeVar, Union, Void,
-    SimpleVector, AbstractArray, DenseArray,
+    SimpleVector, AbstractArray, DenseArray, NamedTuple,
     # special objects
     Function, CodeInfo, Method, MethodTable, TypeMapEntry, TypeMapLevel,
     Module, Symbol, Task, Array, WeakRef, VecElement,

base/inference.jl

@@ -493,6 +493,8 @@ end
 const _Type_name = Type.body.name
 isType(@nospecialize t) = isa(t, DataType) && (t::DataType).name === _Type_name
 
+const _NamedTuple_name = NamedTuple.body.body.name
+
 # true if Type is inlineable as constant (is a singleton)
 function isconstType(@nospecialize t)
     isType(t) || return false
@@ -734,6 +736,10 @@ function isdefined_tfunc(args...)
             end
             if 1 <= idx <= a1.ninitialized
                 return Const(true)
+            elseif a1.name === _NamedTuple_name
+                if isleaftype(a1)
+                    return Const(false)
+                end
             elseif idx <= 0 || (!isvatuple(a1) && idx > fieldcount(a1))
                 return Const(false)
             elseif !isvatuple(a1) && isbits(fieldtype(a1, idx))
@@ -771,7 +777,9 @@ add_tfunc(nfields, 1, 1,
             # TODO: remove with deprecation in builtins.c for nfields(::Type)
             isleaftype(x.parameters[1]) && return Const(old_nfields(x.parameters[1]))
         elseif isa(x,DataType) && !x.abstract && !(x.name === Tuple.name && isvatuple(x)) && x !== DataType
-            return Const(length(x.types))
+            if !(x.name === _NamedTuple_name && !isleaftype(x))
+                return Const(length(x.types))
+            end
         end
         return Int
     end, 0)
@@ -1333,6 +1341,10 @@ function getfield_tfunc(@nospecialize(s00), @nospecialize(name))
         end
         return Any
     end
+    if s.name === _NamedTuple_name && !isleaftype(s)
+        # TODO: better approximate inference
+        return Any
+    end
     if isempty(s.types)
         return Bottom
     end
@@ -1416,6 +1428,9 @@ function fieldtype_tfunc(@nospecialize(s0), @nospecialize(name))
     if !isa(u,DataType) || u.abstract
         return Type
     end
+    if u.name === _NamedTuple_name && !isleaftype(u)
+        return Type
+    end
     ftypes = u.types
     if isempty(ftypes)
         return Bottom

base/namedtuple.jl

@@ -0,0 +1,168 @@
+# This file is a part of Julia. License is MIT: https://julialang.org/license
+
+function NamedTuple{names,T}(args::Tuple) where {names, T <: Tuple}
+    if length(args) == length(names)
+        if @generated
+            N = length(names)
+            types = T.parameters
+            Expr(:new, :(NamedTuple{names,T}), Any[ :(convert($(types[i]), args[$i])) for i in 1:N ]...)
+        else
+            N = length(names)
+            NT = NamedTuple{names,T}
+            ccall(:jl_new_structv, Any, (Any, Ptr{Void}, UInt32), NT, collect(Any, T(args)), N)::NT
+        end
+    else
+        throw(ArgumentError("Wrong number of arguments to named tuple constructor."))
+    end
+end
+
+function NamedTuple{names}(args::Tuple) where {names}
+    NamedTuple{names,typeof(args)}(args)
+end
+
+NamedTuple() = NamedTuple{(),Tuple{}}(())
+
+length(t::NamedTuple) = nfields(t)
+start(t::NamedTuple) = 1
+done(t::NamedTuple, iter) = iter > nfields(t)
+next(t::NamedTuple, iter) = (getfield(t, iter), iter + 1)
+endof(t::NamedTuple) = nfields(t)
+getindex(t::NamedTuple, i::Int) = getfield(t, i)
+getindex(t::NamedTuple, i::Symbol) = getfield(t, i)
+indexed_next(t::NamedTuple, i::Int, state) = (getfield(t,i), i+1)
+
+convert(::Type{NamedTuple{names,T}}, nt::NamedTuple{names,T}) where {names,T} = nt
+convert(::Type{NamedTuple{names}}, nt::NamedTuple{names}) where {names} = nt
+
+function convert(::Type{NamedTuple{names,T}}, nt::NamedTuple{names}) where {names,T}
+    NamedTuple{names,T}(T(nt))
+end
+
+function show(io::IO, t::NamedTuple)
+    n = nfields(t)
+    if n == 0
+        print(io, "NamedTuple()")
+    else
+        print(io, "(")
+        for i = 1:n
+            print(io, fieldname(typeof(t),i), " = "); show(io, getfield(t,i))
+            if n == 1
+                print(io, ",")
+            elseif i < n
+                print(io, ", ")
+            end
+        end
+        print(io, ")")
+    end
+end
+
+eltype(::Type{NamedTuple{names,T}}) where {names,T} = eltype(T)
+
+==(a::NamedTuple{n}, b::NamedTuple{n}) where {n} = Tuple(a) == Tuple(b)
+==(a::NamedTuple, b::NamedTuple) = false
+
+isequal(a::NamedTuple{n}, b::NamedTuple{n}) where {n} = isequal(Tuple(a), Tuple(b))
+isequal(a::NamedTuple, b::NamedTuple) = false
+
+_nt_names(::NamedTuple{names}) where {names} = names
+_nt_names(::Type{T}) where {names,T<:NamedTuple{names}} = names
+
+hash(x::NamedTuple, h::UInt) = xor(object_id(_nt_names(x)), hash(Tuple(x), h))
+
+isless(a::NamedTuple{n}, b::NamedTuple{n}) where {n} = isless(Tuple(a), Tuple(b))
+# TODO: case where one argument's names are a prefix of the other's
+
+same_names(::NamedTuple{names}...) where {names} = true
+same_names(::NamedTuple...) = false
+
+function map(f, nt::NamedTuple{names}, nts::NamedTuple...) where names
+    if !same_names(nt, nts...)
+        throw(ArgumentError("Named tuple names do not match."))
+    end
+    # this method makes sure we don't define a map(f) method
+    NT = NamedTuple{names}
+    if @generated
+        N = length(names)
+        M = length(nts)
+        args = Expr[:(f($(Expr[:(getfield(nt, $j)), (:(getfield(nts[$i], $j)) for i = 1:M)...]...))) for j = 1:N]
+        :( NT(($(args...),)) )
+    else
+        NT(map(f, map(Tuple, (nt, nts...))...))
+    end
+end
+
+# a version of `in` for the older world these generated functions run in
+@pure function sym_in(x::Symbol, itr::Tuple{Vararg{Symbol}})
+    for y in itr
+        y === x && return true
+    end
+    return false
+end
+
+@pure function merge_names(an::Tuple{Vararg{Symbol}}, bn::Tuple{Vararg{Symbol}})
+    names = Symbol[an...]
+    for n in bn
+        if !sym_in(n, an)
+            push!(names, n)
+        end
+    end
+    (names...,)
+end
+
+@pure function merge_types(names::Tuple{Vararg{Symbol}}, a::Type{<:NamedTuple}, b::Type{<:NamedTuple})
+    bn = _nt_names(b)
+    Tuple{Any[ fieldtype(sym_in(n, bn) ? b : a, n) for n in names ]...}
+end
+
+"""
+    merge(a::NamedTuple, b::NamedTuple)
+
+Construct a new named tuple by merging two existing ones.
+The order of fields in `a` is preserved, but values are taken from matching
+fields in `b`. Fields present only in `b` are appended at the end.
+
+```jldoctest
+julia> merge((a=1, b=2, c=3), (b=4, d=5))
+(a = 1, b = 4, c = 3, d = 5)
+```
+"""
+function merge(a::NamedTuple{an}, b::NamedTuple{bn}) where {an, bn}
+    if @generated
+        names = merge_names(an, bn)
+        types = merge_types(names, a, b)
+        vals = Any[ :(getfield($(sym_in(n, bn) ? :b : :a), $(Expr(:quote, n)))) for n in names ]
+        :( NamedTuple{$names,$types}(($(vals...),)) )
+    else
+        names = merge_names(an, bn)
+        types = merge_types(names, typeof(a), typeof(b))
+        NamedTuple{names,types}(map(n->getfield(sym_in(n, bn) ? b : a, n), names))
+    end
+end
+
+merge(a::NamedTuple{()}, b::NamedTuple) = b
+
+"""
+    merge(a::NamedTuple, iterable)
+
+Interpret an iterable of key-value pairs as a named tuple, and perform a merge.
+
+```jldoctest
+julia> merge((a=1, b=2, c=3), [:b=>4, :d=>5])
+(a = 1, b = 4, c = 3, d = 5)
+```
+"""
+function merge(a::NamedTuple, itr)
+    names = Symbol[]
+    vals = Any[]
+    for (k,v) in itr
+        push!(names, k)
+        push!(vals, v)
+    end
+    merge(a, NamedTuple{(names...)}((vals...,)))
+end
+
+keys(nt::NamedTuple{names}) where {names} = names
+values(nt::NamedTuple) = Tuple(nt)
+haskey(nt::NamedTuple, key::Union{Integer, Symbol}) = isdefined(nt, key)
+get(nt::NamedTuple, key::Union{Integer, Symbol}, default) = haskey(nt, key) ? getfield(nt, key) : default
+get(f::Callable, nt::NamedTuple, key::Union{Integer, Symbol}) = haskey(nt, key) ? getfield(nt, key) : f()

base/reflection.jl

@@ -127,12 +127,14 @@ julia> fieldname(SparseMatrixCSC, 5)
 ```
 """
 function fieldname(t::DataType, i::Integer)
-    n_fields = length(t.name.names)
+    names = isdefined(t, :names) ? t.names : t.name.names
+    n_fields = length(names)
     field_label = n_fields == 1 ? "field" : "fields"
     i > n_fields && throw(ArgumentError("Cannot access field $i since type $t only has $n_fields $field_label."))
     i < 1 && throw(ArgumentError("Field numbers must be positive integers. $i is invalid."))
-    return t.name.names[i]::Symbol
+    return names[i]::Symbol
 end
+
 fieldname(t::UnionAll, i::Integer) = fieldname(unwrap_unionall(t), i)
 fieldname(t::Type{<:Tuple}, i::Integer) =
     i < 1 || i > fieldcount(t) ? throw(BoundsError(t, i)) : Int(i)

base/sysimg.jl

@@ -142,6 +142,8 @@ Vector(m::Integer) = Array{Any,1}(Int(m))
 Matrix{T}(m::Integer, n::Integer) where {T} = Matrix{T}(Int(m), Int(n))
 Matrix(m::Integer, n::Integer) = Matrix{Any}(Int(m), Int(n))
 
+include("namedtuple.jl")
+
 # numeric operations
 include("hashing.jl")
 include("rounding.jl")

doc/src/manual/functions.md

@@ -221,6 +221,48 @@ A zero-argument anonymous function is written as `()->3`. The idea of a function
 may seem strange, but is useful for "delaying" a computation. In this usage, a block of code is
 wrapped in a zero-argument function, which is later invoked by calling it as `f`.
 
+## Tuples
+
+Julia has a built-in data structure called a *tuple* that is closely related to function
+arguments and return values.
+A tuple is a fixed-length container that can hold any values, but cannot be modified
+(it is *immutable*).
+Tuples are constructed with commas and parentheses, and can be accessed via indexing:
+
+```jldoctest
+julia> (1, 1+1)
+(1, 2)
+
+julia> (1,)
+(1,)
+
+julia> x = (0.0, "hello", 6*7)
+(0.0, "hello", 42)
+
+julia> x[2]
+"hello"
+```
+
+Notice that a length-1 tuple must be written with a comma, `(1,)`, since `(1)` would just
+be a parenthesized value.
+`()` represents the empty (length-0) tuple.
+
+## Named Tuples
+
+The components of tuples can optionally be named, in which case a *named tuple* is
+constructed:
+
+```jldoctest
+julia> x = (a=1, b=1+1)
+(a = 1, b = 2)
+
+julia> x.a
+1
+```
+
+Named tuples are very similar to tuples, except that fields can additionally be accessed by name
+using dot syntax (`x.a`).
+
 ## Multiple Return Values
 
 In Julia, one returns a tuple of values to simulate returning multiple values. However, tuples
@@ -320,7 +362,7 @@ In all these cases, `x` is bound to a tuple of the trailing values passed to `ba
 It is possible to constrain the number of values passed as a variable argument; this will be discussed
 later in [Parametrically-constrained Varargs methods](@ref).
 
-On the flip side, it is often handy to "splice" the values contained in an iterable collection
+On the flip side, it is often handy to "splat" the values contained in an iterable collection
 into a function call as individual arguments. To do this, one also uses `...` but in the function
 call instead:
 
@@ -349,7 +391,7 @@ julia> bar(x...)
 (1, 2, (3, 4))
 ```
 
-Furthermore, the iterable object spliced into a function call need not be a tuple:
+Furthermore, the iterable object splatted into a function call need not be a tuple:
 
 ```jldoctest barfunc
 julia> x = [3,4]
@@ -371,7 +413,7 @@ julia> bar(x...)
 (1, 2, (3, 4))
 ```
 
-Also, the function that arguments are spliced into need not be a varargs function (although it
+Also, the function that arguments are splatted into need not be a varargs function (although it
 often is):
 
 ```jldoctest
@@ -397,7 +439,7 @@ Closest candidates are:
   baz(::Any, ::Any) at none:1
 ```
 
-As you can see, if the wrong number of elements are in the spliced container, then the function
+As you can see, if the wrong number of elements are in the splatted container, then the function
 call will fail, just as it would if too many arguments were given explicitly.
 
 ## Optional Arguments

doc/src/manual/types.md

@@ -905,6 +905,31 @@ used to represent the arguments accepted by varargs methods (see [Varargs Functi
 The type `Vararg{T,N}` corresponds to exactly `N` elements of type `T`.  `NTuple{N,T}` is a convenient
 alias for `Tuple{Vararg{T,N}}`, i.e. a tuple type containing exactly `N` elements of type `T`.
 
+### Named Tuple Types
+
+Named tuples are instances of the `NamedTuple` type, which has two parameters: a tuple of
+symbols giving the field names, and a tuple type giving the field types.
+
+```jldoctest
+julia> typeof((a=1,b="hello"))
+NamedTuple{(:a, :b),Tuple{Int64,String}}
+```
+
+A `NamedTuple` type can be used as a constructor, accepting a single tuple argument.
+The constructed `NamedTuple` type can be either a concrete type, with both parameters specified,
+or a type that specifies only field names:
+
+```jldoctest
+julia> NamedTuple{(:a, :b),Tuple{Float32, String}}((1,""))
+(a = 1.0f0, b = "")
+
+julia> NamedTuple{(:a, :b)}((1,""))
+(a = 1, b = "")
+```
+
+If field types are specified, the arguments are converted. Otherwise the types of the arguments
+are used directly.
+
 #### [Singleton Types](@id man-singleton-types)
 
 There is a special kind of abstract parametric type that must be mentioned here: singleton types.

src/ast.scm

@@ -153,6 +153,7 @@
 ;; predicates and accessors
 
 (define (quoted? e) (memq (car e) '(quote top core globalref outerref line break inert meta)))
+(define (quotify e) `',e)
 
 (define (lam:args x) (cadr x))
 (define (lam:vars x) (llist-vars (lam:args x)))

src/builtins.c

@@ -1295,6 +1295,7 @@ void jl_init_primitives(void)
     add_builtin("QuoteNode", (jl_value_t*)jl_quotenode_type);
     add_builtin("NewvarNode", (jl_value_t*)jl_newvarnode_type);
     add_builtin("GlobalRef", (jl_value_t*)jl_globalref_type);
+    add_builtin("NamedTuple", (jl_value_t*)jl_namedtuple_type);
 
     add_builtin("Bool", (jl_value_t*)jl_bool_type);
     add_builtin("UInt8", (jl_value_t*)jl_uint8_type);