add typemap filtering option for Union{}

vtjnash · vtjnash · commit 67aa46273ac4 · 2023-04-23T20:42:39.000-04:00
Based on the new morespecific rule for Union{} and method definitions of the specific form `f(..., Type{Union{}}, Vararg)`. If a method definition exists with that specific form, the intersection visitor will ignore all intersections that have that as their only result, saving significant effort when working with lookups involving `Type{<:T}` (which usually ended up mostly ambiguous anyways). Fixes: #33780 This pattern turns out to have still to been making package loading slow. We could keep adding methods following the ambiguity pattern #46000 for the couple specific functions that need it (constructor, eltype, IteratorEltype, IteratorSize, and maybe a couple others) so the internals can detect those and optimize functions that have that method pair. But it seems somewhat odd, convoluted, and non-obvious behavior there. Instead, this breaks all ambiguities in which Union{} is present explicitly in favor of the method with Union{}. This means that when computing method matches, as soon as we see one method definition with Union{}, we can record that the method is the only possible match for that slot. This, in essence, permits creating a rule for dispatch that a TypeVar lower bound must be strictly a supertype of Union{}, but this creates it at the function level, instead of expecting the user to add it to every TypeVar they use to define methods. This also lets us improve the error message for these cases (generally they should error to avoid polluting the inference result), since we can be assured this method will be called, and not result in an ambiguous MethodError instead! Reverts the functional change of #46000
diff --git a/base/abstractarray.jl b/base/abstractarray.jl
@@ -183,11 +183,13 @@ CartesianIndex{2}
      For arrays, this function requires at least Julia 1.2.
 """
 keytype(a::AbstractArray) = keytype(typeof(a))
+keytype(::Type{Union{}}, slurp...) = eltype(Union{})
 
 keytype(A::Type{<:AbstractArray}) = CartesianIndex{ndims(A)}
 keytype(A::Type{<:AbstractVector}) = Int
 
 valtype(a::AbstractArray) = valtype(typeof(a))
+valtype(::Type{Union{}}, slurp...) = eltype(Union{})
 
 """
     valtype(T::Type{<:AbstractArray})
@@ -232,7 +234,7 @@ UInt8
 ```
 """
 eltype(::Type) = Any
-eltype(::Type{Bottom}) = throw(ArgumentError("Union{} does not have elements"))
+eltype(::Type{Bottom}, slurp...) = throw(ArgumentError("Union{} does not have elements"))
 eltype(x) = eltype(typeof(x))
 eltype(::Type{<:AbstractArray{E}}) where {E} = @isdefined(E) ? E : Any
 
@@ -268,6 +270,7 @@ julia> ndims(A)
 """
 ndims(::AbstractArray{T,N}) where {T,N} = N
 ndims(::Type{<:AbstractArray{<:Any,N}}) where {N} = N
+ndims(::Type{Union{}}, slurp...) = throw(ArgumentError("Union{} does not have elements"))
 
 """
     length(collection) -> Integer
diff --git a/base/array.jl b/base/array.jl
@@ -252,7 +252,10 @@ function bitsunionsize(u::Union)
     return sz
 end
 
+# Deprecate this, as it seems to have no documented meaning and is unused here,
+# but is frequently accessed in packages
 elsize(@nospecialize _::Type{A}) where {T,A<:Array{T}} = aligned_sizeof(T)
+elsize(::Type{Union{}}, slurp...) = 0
 sizeof(a::Array) = Core.sizeof(a)
 
 function isassigned(a::Array, i::Int...)
diff --git a/base/arrayshow.jl b/base/arrayshow.jl
@@ -540,10 +540,12 @@ end
 # returning Any, as this would cause incorrect printing in e.g. `Vector[Any[1]]`,
 # because eltype(Vector) == Any so `Any` wouldn't be printed in `Any[1]`)
 typeinfo_eltype(typeinfo) = nothing # element type not precisely known
+typeinfo_eltype(typeinfo::Type{Union{}}, slurp...) = nothing
 typeinfo_eltype(typeinfo::Type{<:AbstractArray{T}}) where {T} = eltype(typeinfo)
 typeinfo_eltype(typeinfo::Type{<:AbstractDict{K,V}}) where {K,V} = eltype(typeinfo)
 typeinfo_eltype(typeinfo::Type{<:AbstractSet{T}}) where {T} = eltype(typeinfo)
 
+
 # types that can be parsed back accurately from their un-decorated representations
 function typeinfo_implicit(@nospecialize(T))
     if T === Float64 || T === Int || T === Char || T === String || T === Symbol ||
diff --git a/base/boot.jl b/base/boot.jl
@@ -258,9 +258,17 @@ UnionAll(v::TypeVar, @nospecialize(t)) = ccall(:jl_type_unionall, Any, (Any, Any
 
 const Vararg = ccall(:jl_toplevel_eval_in, Any, (Any, Any), Core, _expr(:new, TypeofVararg))
 
-# let the compiler assume that calling Union{} as a constructor does not need
-# to be considered ever (which comes up often as Type{<:T})
-Union{}(a...) = throw(MethodError(Union{}, a))
+# dispatch token indicating a kwarg (keyword sorter) call
+function kwcall end
+# deprecated internal functions:
+kwfunc(@nospecialize(f)) = kwcall
+kwftype(@nospecialize(t)) = typeof(kwcall)
+
+# Let the compiler assume that calling Union{} as a constructor does not need
+# to be considered ever (which comes up often as Type{<:T} inference, and
+# occasionally in user code from eltype).
+Union{}(a...) = throw(ArgumentError("cannot construct a value of type Union{} for return result"))
+kwcall(kwargs, ::Type{Union{}}, a...) = Union{}(a...)
 
 Expr(@nospecialize args...) = _expr(args...)
 
@@ -369,12 +377,6 @@ include(m::Module, fname::String) = ccall(:jl_load_, Any, (Any, Any), m, fname)
 
 eval(m::Module, @nospecialize(e)) = ccall(:jl_toplevel_eval_in, Any, (Any, Any), m, e)
 
-# dispatch token indicating a kwarg (keyword sorter) call
-function kwcall end
-# deprecated internal functions:
-kwfunc(@nospecialize(f)) = kwcall
-kwftype(@nospecialize(t)) = typeof(kwcall)
-
 mutable struct Box
     contents::Any
     Box(@nospecialize(x)) = new(x)
diff --git a/base/broadcast.jl b/base/broadcast.jl
@@ -34,6 +34,9 @@ that you may be able to leverage; see the
 """
 abstract type BroadcastStyle end
 
+struct Unknown <: BroadcastStyle end
+BroadcastStyle(::Type{Union{}}, slurp...) = Unknown()  # ambiguity resolution
+
 """
 `Broadcast.Style{C}()` defines a [`BroadcastStyle`](@ref) signaling through the type
 parameter `C`. You can use this as an alternative to creating custom subtypes of `BroadcastStyle`,
@@ -45,9 +48,6 @@ struct Style{T} <: BroadcastStyle end
 
 BroadcastStyle(::Type{<:Tuple}) = Style{Tuple}()
 
-struct Unknown <: BroadcastStyle end
-BroadcastStyle(::Type{Union{}}) = Unknown()  # ambiguity resolution
-
 """
 `Broadcast.AbstractArrayStyle{N} <: BroadcastStyle` is the abstract supertype for any style
 associated with an `AbstractArray` type.
diff --git a/base/complex.jl b/base/complex.jl
@@ -120,6 +120,7 @@ Float64
 real(T::Type) = typeof(real(zero(T)))
 real(::Type{T}) where {T<:Real} = T
 real(C::Type{<:Complex}) = fieldtype(C, 1)
+real(::Type{Union{}}, slurp...) = Union{}(im)
 
 """
     isreal(x) -> Bool
diff --git a/base/essentials.jl b/base/essentials.jl
@@ -311,7 +311,7 @@ See also: [`round`](@ref), [`trunc`](@ref), [`oftype`](@ref), [`reinterpret`](@r
 function convert end
 
 # ensure this is never ambiguous, and therefore fast for lookup
-convert(T::Type{Union{}}, x) = throw(MethodError(convert, (T, x)))
+convert(T::Type{Union{}}, x...) = throw(ArgumentError("cannot convert a value to Union{} for assignment"))
 
 convert(::Type{Type}, x::Type) = x # the ssair optimizer is strongly dependent on this method existing to avoid over-specialization
                                    # in the absence of inlining-enabled
@@ -535,6 +535,7 @@ Neither `convert` nor `cconvert` should take a Julia object and turn it into a `
 function cconvert end
 
 cconvert(T::Type, x) = x isa T ? x : convert(T, x) # do the conversion eagerly in most cases
+cconvert(::Type{Union{}}, x...) = convert(Union{}, x...)
 cconvert(::Type{<:Ptr}, x) = x # but defer the conversion to Ptr to unsafe_convert
 unsafe_convert(::Type{T}, x::T) where {T} = x # unsafe_convert (like convert) defaults to assuming the convert occurred
 unsafe_convert(::Type{T}, x::T) where {T<:Ptr} = x  # to resolve ambiguity with the next method
diff --git a/base/float.jl b/base/float.jl
@@ -310,6 +310,7 @@ Float64
 """
 float(::Type{T}) where {T<:Number} = typeof(float(zero(T)))
 float(::Type{T}) where {T<:AbstractFloat} = T
+float(::Type{Union{}}, slurp...) = Union{}(0.0)
 
 """
     unsafe_trunc(T, x)
diff --git a/base/generator.jl b/base/generator.jl
@@ -92,13 +92,13 @@ Base.HasLength()
 """
 IteratorSize(x) = IteratorSize(typeof(x))
 IteratorSize(::Type) = HasLength()  # HasLength is the default
+IteratorSize(::Type{Union{}}, slurp...) = throw(ArgumentError("Union{} does not have elements"))
+IteratorSize(::Type{Any}) = SizeUnknown()
 
 IteratorSize(::Type{<:Tuple}) = HasLength()
 IteratorSize(::Type{<:AbstractArray{<:Any,N}})  where {N} = HasShape{N}()
 IteratorSize(::Type{Generator{I,F}}) where {I,F} = IteratorSize(I)
 
-IteratorSize(::Type{Any}) = SizeUnknown()
-
 haslength(iter) = IteratorSize(iter) isa Union{HasShape, HasLength}
 
 abstract type IteratorEltype end
@@ -126,7 +126,7 @@ Base.HasEltype()
 """
 IteratorEltype(x) = IteratorEltype(typeof(x))
 IteratorEltype(::Type) = HasEltype()  # HasEltype is the default
+IteratorEltype(::Type{Union{}}, slurp...) = throw(ArgumentError("Union{} does not have elements"))
+IteratorEltype(::Type{Any}) = EltypeUnknown()
 
 IteratorEltype(::Type{Generator{I,T}}) where {I,T} = EltypeUnknown()
-
-IteratorEltype(::Type{Any}) = EltypeUnknown()
diff --git a/base/indices.jl b/base/indices.jl
@@ -92,7 +92,7 @@ particular, [`eachindex`](@ref) creates an iterator whose type depends
 on the setting of this trait.
 """
 IndexStyle(A::AbstractArray) = IndexStyle(typeof(A))
-IndexStyle(::Type{Union{}}) = IndexLinear()
+IndexStyle(::Type{Union{}}, slurp...) = IndexLinear()
 IndexStyle(::Type{<:AbstractArray}) = IndexCartesian()
 IndexStyle(::Type{<:Array}) = IndexLinear()
 IndexStyle(::Type{<:AbstractRange}) = IndexLinear()
diff --git a/base/io.jl b/base/io.jl
@@ -219,6 +219,8 @@ julia> read(io, String)
 ```
 """
 read(stream, t)
+read(stream, ::Type{Union{}}, slurp...; kwargs...) = error("cannot read a value of type Union{}")
+
 
 """
     write(io::IO, x)
diff --git a/base/iterators.jl b/base/iterators.jl
@@ -1170,6 +1170,7 @@ IteratorEltype(::Type{Flatten{Tuple{}}}) = IteratorEltype(Tuple{})
 _flatteneltype(I, ::HasEltype) = IteratorEltype(eltype(I))
 _flatteneltype(I, et) = EltypeUnknown()
 
+flatten_iteratorsize(::Union{HasShape, HasLength}, ::Type{Union{}}, slurp...) = HasLength() # length==0
 flatten_iteratorsize(::Union{HasShape, HasLength}, ::Type{<:NTuple{N,Any}}) where {N} = HasLength()
 flatten_iteratorsize(::Union{HasShape, HasLength}, ::Type{<:Tuple}) = SizeUnknown()
 flatten_iteratorsize(::Union{HasShape, HasLength}, ::Type{<:Number}) = HasLength()
@@ -1181,6 +1182,7 @@ _flatten_iteratorsize(sz, ::HasEltype, ::Type{Tuple{}}) = HasLength()
 
 IteratorSize(::Type{Flatten{I}}) where {I} = _flatten_iteratorsize(IteratorSize(I), IteratorEltype(I), I)
 
+flatten_length(f, T::Type{Union{}}, slurp...) = 0
 function flatten_length(f, T::Type{<:NTuple{N,Any}}) where {N}
     return N * length(f.it)
 end
diff --git a/base/missing.jl b/base/missing.jl
@@ -41,6 +41,7 @@ nonmissingtype(::Type{T}) where {T} = typesplit(T, Missing)
 function nonmissingtype_checked(T::Type)
     R = nonmissingtype(T)
     R >: T && error("could not compute non-missing type")
+    R <: Union{} && error("cannot convert a value to missing for assignment")
     return R
 end
 
@@ -69,7 +70,6 @@ convert(::Type{T}, x::T) where {T>:Union{Missing, Nothing}} = x
 convert(::Type{T}, x) where {T>:Missing} = convert(nonmissingtype_checked(T), x)
 convert(::Type{T}, x) where {T>:Union{Missing, Nothing}} = convert(nonmissingtype_checked(nonnothingtype_checked(T)), x)
 
-
 # Comparison operators
 ==(::Missing, ::Missing) = missing
 ==(::Missing, ::Any) = missing
diff --git a/base/number.jl b/base/number.jl
@@ -307,6 +307,7 @@ julia> zero(rand(2,2))
 """
 zero(x::Number) = oftype(x,0)
 zero(::Type{T}) where {T<:Number} = convert(T,0)
+zero(::Type{Union{}}, slurp...) = Union{}(0)
 
 """
     one(x)
@@ -345,6 +346,7 @@ julia> import Dates; one(Dates.Day(1))
 """
 one(::Type{T}) where {T<:Number} = convert(T,1)
 one(x::T) where {T<:Number} = one(T)
+one(::Type{Union{}}, slurp...) = Union{}(1)
 # note that convert(T, 1) should throw an error if T is dimensionful,
 # so this fallback definition should be okay.
 
@@ -368,6 +370,7 @@ julia> import Dates; oneunit(Dates.Day)
 """
 oneunit(x::T) where {T} = T(one(x))
 oneunit(::Type{T}) where {T} = T(one(T))
+oneunit(::Type{Union{}}, slurp...) = Union{}(1)
 
 """
     big(T::Type)
@@ -388,3 +391,4 @@ Complex{BigInt}
 ```
 """
 big(::Type{T}) where {T<:Number} = typeof(big(zero(T)))
+big(::Type{Union{}}, slurp...) = Union{}(0)
diff --git a/base/operators.jl b/base/operators.jl
@@ -888,6 +888,7 @@ julia> widen(1.5f0)
 """
 widen(x::T) where {T} = convert(widen(T), x)
 widen(x::Type{T}) where {T} = throw(MethodError(widen, (T,)))
+widen(x::Type{Union{}}, slurp...) = throw(MethodError(widen, (Union{},)))
 
 # function pipelining
 
diff --git a/base/parse.jl b/base/parse.jl
@@ -36,6 +36,7 @@ julia> parse(Complex{Float64}, "3.2e-1 + 4.5im")
 ```
 """
 parse(T::Type, str; base = Int)
+parse(::Type{Union{}}, slurp...; kwargs...) = error("cannot parse a value as Union{}")
 
 function parse(::Type{T}, c::AbstractChar; base::Integer = 10) where T<:Integer
     a::Int = (base <= 36 ? 10 : 36)
@@ -251,6 +252,7 @@ function parse(::Type{T}, s::AbstractString; base::Union{Nothing,Integer} = noth
     convert(T, tryparse_internal(T, s, firstindex(s), lastindex(s),
                                  base===nothing ? 0 : check_valid_base(base), true))
 end
+tryparse(::Type{Union{}}, slurp...; kwargs...) = error("cannot parse a value as Union{}")
 
 ## string to float functions ##
 
diff --git a/base/promotion.jl b/base/promotion.jl
@@ -323,6 +323,12 @@ it for new types as appropriate.
 function promote_rule end
 
 promote_rule(::Type, ::Type) = Bottom
+# Define some methods to avoid needing to enumerate unrelated possibilities when presented
+# with Type{<:T}, and return a value in general accordance with the result given by promote_type
+promote_rule(::Type{Bottom}, slurp...) = Bottom
+promote_rule(::Type{Bottom}, ::Type{Bottom}, slurp...) = Bottom # not strictly necessary, since the next method would match unambiguously anyways
+promote_rule(::Type{Bottom}, ::Type{T}, slurp...) where {T} = T
+promote_rule(::Type{T}, ::Type{Bottom}, slurp...) where {T} = T
 
 promote_result(::Type,::Type,::Type{T},::Type{S}) where {T,S} = (@inline; promote_type(T,S))
 # If no promote_rule is defined, both directions give Bottom. In that
diff --git a/base/some.jl b/base/some.jl
@@ -29,6 +29,7 @@ end
 function nonnothingtype_checked(T::Type)
     R = nonnothingtype(T)
     R >: T && error("could not compute non-nothing type")
+    R <: Union{} && error("cannot convert a value to nothing for assignment")
     return R
 end
 
diff --git a/base/traits.jl b/base/traits.jl
@@ -11,7 +11,7 @@ OrderStyle(::Type{<:Real}) = Ordered()
 OrderStyle(::Type{<:AbstractString}) = Ordered()
 OrderStyle(::Type{Symbol}) = Ordered()
 OrderStyle(::Type{<:Any}) = Unordered()
-OrderStyle(::Type{Union{}}) = Ordered()
+OrderStyle(::Type{Union{}}, slurp...) = Ordered()
 
 # trait for objects that support arithmetic
 abstract type ArithmeticStyle end
@@ -23,6 +23,7 @@ ArithmeticStyle(instance) = ArithmeticStyle(typeof(instance))
 ArithmeticStyle(::Type{<:AbstractFloat}) = ArithmeticRounds()
 ArithmeticStyle(::Type{<:Integer}) = ArithmeticWraps()
 ArithmeticStyle(::Type{<:Any}) = ArithmeticUnknown()
+ArithmeticStyle(::Type{Union{}}, slurp...) = ArithmeticUnknown()
 
 # trait for objects that support ranges with regular step
 """
@@ -58,5 +59,6 @@ ranges with an element type which is a subtype of `Integer`.
 abstract type RangeStepStyle end
 struct RangeStepRegular   <: RangeStepStyle end # range with regular step
 struct RangeStepIrregular <: RangeStepStyle end # range with rounding error
+RangeStepStyle(::Type{Union{}}, slurp...) = RangeStepIrregular()
 
 RangeStepStyle(instance) = RangeStepStyle(typeof(instance))
diff --git a/src/gf.c b/src/gf.c
@@ -1487,6 +1487,8 @@ static int get_intersect_visitor(jl_typemap_entry_t *oldentry, struct typemap_in
         // skip if no world has both active
         // also be careful not to try to scan something from the current dump-reload though
         return 1;
+    // don't need to consider other similar methods if this oldentry will always fully intersect with them and dominates all of them
+    typemap_slurp_search(oldentry, &closure->match);
     jl_method_t *oldmethod = oldentry->func.method;
     if (closure->match.issubty // e.g. jl_subtype(closure->newentry.sig, oldentry->sig)
         && jl_subtype(oldmethod->sig, (jl_value_t*)closure->newentry->sig)) { // e.g. jl_type_equal(closure->newentry->sig, oldentry->sig)
@@ -1511,7 +1513,7 @@ static jl_value_t *get_intersect_matches(jl_typemap_t *defs, jl_typemap_entry_t
         else
             va = NULL;
     }
-    struct matches_env env = {{get_intersect_visitor, (jl_value_t*)type, va,
+    struct matches_env env = {{get_intersect_visitor, (jl_value_t*)type, va, /* .search_slurp = */ 0,
             /* .ti = */ NULL, /* .env = */ jl_emptysvec, /* .issubty = */ 0},
         /* .newentry = */ newentry, /* .shadowed */ NULL, /* .replaced */ NULL};
     JL_GC_PUSH3(&env.match.env, &env.match.ti, &env.shadowed);
@@ -3208,6 +3210,7 @@ struct ml_matches_env {
     int intersections;
     size_t world;
     int lim;
+    int include_ambiguous;
     // results:
     jl_value_t *t; // array of method matches
     size_t min_valid;
@@ -3263,6 +3266,9 @@ static int ml_matches_visitor(jl_typemap_entry_t *ml, struct typemap_intersectio
             return 0;
         closure->lim--;
     }
+    // don't need to consider other similar methods if this ml will always fully intersect with them and dominates all of them
+    if (!closure->include_ambiguous || closure->lim != -1)
+        typemap_slurp_search(ml, &closure->match);
     closure->matc = make_method_match((jl_tupletype_t*)closure->match.ti,
         closure->match.env, meth,
         closure->match.issubty ? FULLY_COVERS : NOT_FULLY_COVERS);
@@ -3277,9 +3283,10 @@ static int ml_matches_visitor(jl_typemap_entry_t *ml, struct typemap_intersectio
     return 1;
 }
 
-static int ml_mtable_visitor(jl_methtable_t *mt, void *env)
+static int ml_mtable_visitor(jl_methtable_t *mt, void *closure0)
 {
-    return jl_typemap_intersection_visitor(jl_atomic_load_relaxed(&mt->defs), jl_cachearg_offset(mt), (struct typemap_intersection_env*)env);
+    struct typemap_intersection_env* env = (struct typemap_intersection_env*)closure0;
+    return jl_typemap_intersection_visitor(jl_atomic_load_relaxed(&mt->defs), jl_cachearg_offset(mt), env);
 }
 
 // This is the collect form of calling jl_typemap_intersection_visitor
@@ -3311,9 +3318,9 @@ static jl_value_t *ml_matches(jl_methtable_t *mt,
         else
             va = NULL;
     }
-    struct ml_matches_env env = {{ml_matches_visitor, (jl_value_t*)type, va,
+    struct ml_matches_env env = {{ml_matches_visitor, (jl_value_t*)type, va, /* .search_slurp = */ 0,
             /* .ti = */ NULL, /* .env = */ jl_emptysvec, /* .issubty = */ 0},
-        intersections, world, lim, /* .t = */ jl_an_empty_vec_any,
+        intersections, world, lim, include_ambiguous, /* .t = */ jl_an_empty_vec_any,
         /* .min_valid = */ *min_valid, /* .max_valid = */ *max_valid, /* .matc = */ NULL};
     struct jl_typemap_assoc search = {(jl_value_t*)type, world, jl_emptysvec, 1, ~(size_t)0};
     jl_value_t *isect2 = NULL;
@@ -3377,7 +3384,7 @@ static jl_value_t *ml_matches(jl_methtable_t *mt,
                 return env.t;
             }
         }
-        if (!jl_typemap_intersection_visitor(jl_atomic_load_relaxed(&mt->defs), jl_cachearg_offset(mt), &env.match)) {
+        if (!ml_mtable_visitor(mt, &env.match)) {
             JL_GC_POP();
             return jl_nothing;
         }
diff --git a/src/jltypes.c b/src/jltypes.c
@@ -1708,7 +1708,7 @@ static void check_datatype_parameters(jl_typename_t *tn, jl_value_t **params, si
     JL_GC_POP();
 }
 
-static jl_value_t *extract_wrapper(jl_value_t *t JL_PROPAGATES_ROOT) JL_GLOBALLY_ROOTED
+jl_value_t *extract_wrapper(jl_value_t *t JL_PROPAGATES_ROOT) JL_GLOBALLY_ROOTED
 {
     t = jl_unwrap_unionall(t);
     if (jl_is_datatype(t))
diff --git a/src/julia_internal.h b/src/julia_internal.h
diff --git a/src/subtype.c b/src/subtype.c
diff --git a/src/typemap.c b/src/typemap.c
diff --git a/test/abstractarray.jl b/test/abstractarray.jl
diff --git a/test/ambiguous.jl b/test/ambiguous.jl
diff --git a/test/core.jl b/test/core.jl
diff --git a/test/missing.jl b/test/missing.jl
diff --git a/test/some.jl b/test/some.jl
