Merge pull request #46663 from JuliaLang/backports-release-1.8

Backports for Julia 1.8.2

.github/workflows/statuses.yml

@@ -48,7 +48,6 @@ jobs:
       - run: |
           declare -a CONTEXT_LIST=(
                 "buildbot/tester_freebsd64"
-                "buildbot/tester_macos64"
                 "buildbot/tester_win32"
                 "buildbot/tester_win64"
                 )

base/array.jl

@@ -2308,7 +2308,7 @@ findall(testf::Function, A) = collect(first(p) for p in pairs(A) if testf(last(p
 
 # Broadcasting is much faster for small testf, and computing
 # integer indices from logical index using findall has a negligible cost
-findall(testf::Function, A::AbstractArray) = findall(testf.(A))
+findall(testf::F, A::AbstractArray) where {F<:Function} = findall(testf.(A))
 
 """
     findall(A)

base/binaryplatforms.jl

@@ -1027,7 +1027,7 @@ function platforms_match(a::AbstractPlatform, b::AbstractPlatform)
 
         # Call the comparator, passing in which objects requested this comparison (one, the other, or both)
         # For some comparators this doesn't matter, but for non-symmetrical comparisons, it does.
-        if !comparator(ak, bk, a_comp == comparator, b_comp == comparator)
+        if !(comparator(ak, bk, a_comp == comparator, b_comp == comparator)::Bool)
             return false
         end
     end

base/compiler/abstractinterpretation.jl

@@ -111,7 +111,7 @@ function abstract_call_gf_by_type(interp::AbstractInterpreter, @nospecialize(f),
                 result = abstract_call_method(interp, method, sig_n, svec(), multiple_matches, sv)
                 rt = result.rt
                 edge = result.edge
-                edge !== nothing && push!(edges, edge)
+                edge === nothing || push!(edges, edge)
                 this_argtypes = isa(matches, MethodMatches) ? argtypes : matches.applicable_argtypes[i]
                 this_arginfo = ArgInfo(fargs, this_argtypes)
                 const_call_result = abstract_call_method_with_const_args(interp, result,
@@ -136,25 +136,11 @@ function abstract_call_gf_by_type(interp::AbstractInterpreter, @nospecialize(f),
             this_conditional = ignorelimited(this_rt)
             this_rt = widenwrappedconditional(this_rt)
         else
-            if infer_compilation_signature(interp)
-                # Also infer the compilation signature for this method, so it's available
-                # to the compiler in case it ends up needing it (which is likely).
-                csig = get_compileable_sig(method, sig, match.sparams)
-                if csig !== nothing && csig !== sig
-                    # The result of this inference is not directly used, so temporarily empty
-                    # the use set for the current SSA value.
-                    saved_uses = sv.ssavalue_uses[sv.currpc]
-                    sv.ssavalue_uses[sv.currpc] = empty_bitset
-                    abstract_call_method(interp, method, csig, match.sparams, multiple_matches, sv)
-                    sv.ssavalue_uses[sv.currpc] = saved_uses
-                end
-            end
-
             result = abstract_call_method(interp, method, sig, match.sparams, multiple_matches, sv)
             this_conditional = ignorelimited(result.rt)
             this_rt = widenwrappedconditional(result.rt)
             edge = result.edge
-            edge !== nothing && push!(edges, edge)
+            edge === nothing || push!(edges, edge)
             # try constant propagation with argtypes for this match
             # this is in preparation for inlining, or improving the return result
             this_argtypes = isa(matches, MethodMatches) ? argtypes : matches.applicable_argtypes[i]
@@ -214,6 +200,26 @@ function abstract_call_gf_by_type(interp::AbstractInterpreter, @nospecialize(f),
 
     rettype = from_interprocedural!(rettype, sv, arginfo, conditionals)
 
+    # Also considering inferring the compilation signature for this method, so
+    # it is available to the compiler in case it ends up needing it.
+    if infer_compilation_signature(interp) && 1 == seen == napplicable && rettype !== Any && rettype !== Union{} && !is_removable_if_unused(all_effects)
+        match = applicable[1]::MethodMatch
+        method = match.method
+        sig = match.spec_types
+        mi = specialize_method(match; preexisting=true)
+        if mi !== nothing && !const_prop_methodinstance_heuristic(interp, match, mi::MethodInstance, arginfo, sv)
+            csig = get_compileable_sig(method, sig, match.sparams)
+            if csig !== nothing && csig !== sig
+                # The result of this inference is not directly used, so temporarily empty
+                # the use set for the current SSA value.
+                saved_uses = sv.ssavalue_uses[sv.currpc]
+                sv.ssavalue_uses[sv.currpc] = empty_bitset
+                abstract_call_method(interp, method, csig, match.sparams, multiple_matches, sv)
+                sv.ssavalue_uses[sv.currpc] = saved_uses
+            end
+        end
+    end
+
     if call_result_unused(sv) && !(rettype === Bottom)
         add_remark!(interp, sv, "Call result type was widened because the return value is unused")
         # We're mainly only here because the optimizer might want this code,
@@ -765,13 +771,25 @@ function collect_const_args((; argtypes)::ArgInfo)
                     isa(a, Const) ? a.val :
                     isconstType(a) ? (a::DataType).parameters[1] :
                     (a::DataType).instance
-                end for i in 2:length(argtypes) ]
+                end for i = 2:length(argtypes) ]
+end
+
+struct InvokeCall
+    types     # ::Type
+    lookupsig # ::Type
+    InvokeCall(@nospecialize(types), @nospecialize(lookupsig)) = new(types, lookupsig)
 end
 
 function concrete_eval_call(interp::AbstractInterpreter,
-    @nospecialize(f), result::MethodCallResult, arginfo::ArgInfo, sv::InferenceState)
+    @nospecialize(f), result::MethodCallResult, arginfo::ArgInfo, sv::InferenceState,
+    invokecall::Union{Nothing,InvokeCall}=nothing)
     concrete_eval_eligible(interp, f, result, arginfo, sv) || return nothing
     args = collect_const_args(arginfo)
+    if invokecall !== nothing
+        # this call should be `invoke`d, rewrite `args` back now
+        pushfirst!(args, f, invokecall.types)
+        f = invoke
+    end
     world = get_world_counter(interp)
     value = try
         Core._call_in_world_total(world, f, args...)
@@ -811,13 +829,13 @@ struct ConstCallResults
         new(rt, const_result, effects)
 end
 
-function abstract_call_method_with_const_args(interp::AbstractInterpreter, result::MethodCallResult,
-                                              @nospecialize(f), arginfo::ArgInfo, match::MethodMatch,
-                                              sv::InferenceState)
+function abstract_call_method_with_const_args(interp::AbstractInterpreter,
+    result::MethodCallResult, @nospecialize(f), arginfo::ArgInfo, match::MethodMatch,
+    sv::InferenceState, invokecall::Union{Nothing,InvokeCall}=nothing)
     if !const_prop_enabled(interp, sv, match)
         return nothing
     end
-    val = concrete_eval_call(interp, f, result, arginfo, sv)
+    val = concrete_eval_call(interp, f, result, arginfo, sv, invokecall)
     if val !== nothing
         add_backedge!(result.edge, sv)
         return val
@@ -1547,10 +1565,10 @@ function abstract_invoke(interp::AbstractInterpreter, (; fargs, argtypes)::ArgIn
     nargtype isa DataType || return CallMeta(Any, Effects(), false) # other cases are not implemented below
     isdispatchelem(ft) || return CallMeta(Any, Effects(), false) # check that we might not have a subtype of `ft` at runtime, before doing supertype lookup below
     ft = ft::DataType
-    types = rewrap_unionall(Tuple{ft, unwrap_unionall(types).parameters...}, types)::Type
+    lookupsig = rewrap_unionall(Tuple{ft, unwrap_unionall(types).parameters...}, types)::Type
     nargtype = Tuple{ft, nargtype.parameters...}
     argtype = Tuple{ft, argtype.parameters...}
-    match, valid_worlds, overlayed = findsup(types, method_table(interp))
+    match, valid_worlds, overlayed = findsup(lookupsig, method_table(interp))
     match === nothing && return CallMeta(Any, Effects(), false)
     update_valid_age!(sv, valid_worlds)
     method = match.method
@@ -1569,8 +1587,10 @@ function abstract_invoke(interp::AbstractInterpreter, (; fargs, argtypes)::ArgIn
     #     t, a = ti.parameters[i], argtypes′[i]
     #     argtypes′[i] = t ⊑ a ? t : a
     # end
-    const_call_result = abstract_call_method_with_const_args(interp, result,
-        overlayed ? nothing : singleton_type(ft′), arginfo, match, sv)
+    f = overlayed ? nothing : singleton_type(ft′)
+    invokecall = InvokeCall(types, lookupsig)
+    const_call_result = abstract_call_method_with_const_args(interp,
+        result, f, arginfo, match, sv, invokecall)
     const_result = nothing
     if const_call_result !== nothing
         if const_call_result.rt ⊑ rt

base/compiler/methodtable.jl

@@ -44,17 +44,23 @@ struct OverlayMethodTable <: MethodTableView
     mt::Core.MethodTable
 end
 
+struct MethodMatchKey
+    sig # ::Type
+    limit::Int
+    MethodMatchKey(@nospecialize(sig), limit::Int) = new(sig, limit)
+end
+
 """
     struct CachedMethodTable <: MethodTableView
 
 Overlays another method table view with an additional local fast path cache that
 can respond to repeated, identical queries faster than the original method table.
 """
 struct CachedMethodTable{T} <: MethodTableView
-    cache::IdDict{Any, Union{Missing, MethodMatchResult}}
+    cache::IdDict{MethodMatchKey, Union{Missing,MethodMatchResult}}
     table::T
 end
-CachedMethodTable(table::T) where T = CachedMethodTable{T}(IdDict{Any, Union{Missing, MethodMatchResult}}(), table)
+CachedMethodTable(table::T) where T = CachedMethodTable{T}(IdDict{MethodMatchKey, Union{Missing,MethodMatchResult}}(), table)
 
 """
     findall(sig::Type, view::MethodTableView; limit::Int=typemax(Int)) ->
@@ -109,9 +115,11 @@ function findall(@nospecialize(sig::Type), table::CachedMethodTable; limit::Int=
         # as for concrete types, we cache result at on the next level
         return findall(sig, table.table; limit)
     end
-    box = Core.Box(sig)
-    return get!(table.cache, sig) do
-        findall(box.contents, table.table; limit)
+    key = MethodMatchKey(sig, limit)
+    if haskey(table.cache, key)
+        return table.cache[key]
+    else
+        return table.cache[key] = findall(sig, table.table; limit)
     end
 end
 

base/compiler/ssair/inlining.jl

@@ -1066,6 +1066,11 @@ function inline_invoke!(
     return nothing
 end
 
+function invoke_signature(invokesig::Vector{Any})
+    ft, argtyps = widenconst(invokesig[2]), instanceof_tfunc(widenconst(invokesig[3]))[1]
+    return rewrap_unionall(Tuple{ft, unwrap_unionall(argtyps).parameters...}, argtyps)
+end
+
 function narrow_opaque_closure!(ir::IRCode, stmt::Expr, @nospecialize(info), state::InliningState)
     if isa(info, OpaqueClosureCreateInfo)
         lbt = argextype(stmt.args[2], ir)

base/gmp.jl

@@ -667,8 +667,12 @@ function prod(arr::AbstractArray{BigInt})
     # to account for the rounding to limbs in MPZ.mul!
     # (BITS_PER_LIMB-1 would typically be enough, to which we add
     # 1 for the initial multiplication by init=1 in foldl)
-    nbits = GC.@preserve arr sum(arr; init=BITS_PER_LIMB) do x
-        abs(x.size) * BITS_PER_LIMB - leading_zeros(unsafe_load(x.d))
+    nbits = BITS_PER_LIMB
+    for x in arr
+        iszero(x) && return zero(BigInt)
+        xsize = abs(x.size)
+        lz = GC.@preserve x leading_zeros(unsafe_load(x.d, xsize))
+        nbits += xsize * BITS_PER_LIMB - lz
     end
     init = BigInt(; nbits)
     MPZ.set_si!(init, 1)

base/loading.jl

@@ -1587,12 +1587,14 @@ function create_expr_cache(pkg::PkgId, input::String, output::String, concrete_d
     deps_eltype = sprint(show, eltype(concrete_deps); context = :module=>nothing)
     deps = deps_eltype * "[" * join(deps_strs, ",") * "]"
     trace = isassigned(PRECOMPILE_TRACE_COMPILE) ? `--trace-compile=$(PRECOMPILE_TRACE_COMPILE[])` : ``
-    io = open(pipeline(`$(julia_cmd()::Cmd) -O0
-                       --output-ji $output --output-incremental=yes
-                       --startup-file=no --history-file=no --warn-overwrite=yes
-                       --color=$(have_color === nothing ? "auto" : have_color ? "yes" : "no")
-                       $trace
-                       -`, stderr = internal_stderr, stdout = internal_stdout),
+    io = open(pipeline(addenv(`$(julia_cmd()::Cmd) -O0
+                              --output-ji $output --output-incremental=yes
+                              --startup-file=no --history-file=no --warn-overwrite=yes
+                              --color=$(have_color === nothing ? "auto" : have_color ? "yes" : "no")
+                              $trace
+                              -`,
+                              "OPENBLAS_NUM_THREADS" => 1),
+                       stderr = internal_stderr, stdout = internal_stdout),
               "w", stdout)
     # write data over stdin to avoid the (unlikely) case of exceeding max command line size
     write(io.in, """

base/namedtuple.jl

@@ -111,7 +111,8 @@ function NamedTuple{names}(nt::NamedTuple) where {names}
         types = Tuple{(fieldtype(nt, idx[n]) for n in 1:length(idx))...}
         Expr(:new, :(NamedTuple{names, $types}), Any[ :(getfield(nt, $(idx[n]))) for n in 1:length(idx) ]...)
     else
-        types = Tuple{(fieldtype(typeof(nt), names[n]) for n in 1:length(names))...}
+        length_names = length(names)::Integer
+        types = Tuple{(fieldtype(typeof(nt), names[n]) for n in 1:length_names)...}
         NamedTuple{names, types}(map(Fix1(getfield, nt), names))
     end
 end
@@ -318,8 +319,8 @@ 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()
-tail(t::NamedTuple{names}) where names = NamedTuple{tail(names)}(t)
-front(t::NamedTuple{names}) where names = NamedTuple{front(names)}(t)
+tail(t::NamedTuple{names}) where names = NamedTuple{tail(names::Tuple)}(t)
+front(t::NamedTuple{names}) where names = NamedTuple{front(names::Tuple)}(t)
 
 @pure function diff_names(an::Tuple{Vararg{Symbol}}, bn::Tuple{Vararg{Symbol}})
     @nospecialize an bn

base/ordering.jl

@@ -122,7 +122,7 @@ lt(o::Lt,                    a, b) = o.lt(a,b)
 @propagate_inbounds function lt(p::Perm, a::Integer, b::Integer)
     da = p.data[a]
     db = p.data[b]
-    lt(p.order, da, db) | (!lt(p.order, db, da) & (a < b))
+    (lt(p.order, da, db)::Bool) | (!(lt(p.order, db, da)::Bool) & (a < b))
 end
 
 _ord(lt::typeof(isless), by::typeof(identity), order::Ordering) = order

base/permuteddimsarray.jl

@@ -263,11 +263,21 @@ end
     P
 end
 
-function Base._mapreduce_dim(f, op, init::Base._InitialValue, A::PermutedDimsArray, dims::Colon)
+const CommutativeOps = Union{typeof(+),typeof(Base.add_sum),typeof(min),typeof(max),typeof(Base._extrema_rf),typeof(|),typeof(&)}
+
+function Base._mapreduce_dim(f, op::CommutativeOps, init::Base._InitialValue, A::PermutedDimsArray, dims::Colon)
+    Base._mapreduce_dim(f, op, init, parent(A), dims)
+end
+function Base._mapreduce_dim(f::typeof(identity), op::Union{typeof(Base.mul_prod),typeof(*)}, init::Base._InitialValue, A::PermutedDimsArray{<:Union{Real,Complex}}, dims::Colon)
     Base._mapreduce_dim(f, op, init, parent(A), dims)
 end
 
-function Base.mapreducedim!(f, op, B::AbstractArray{T,N}, A::PermutedDimsArray{T,N,perm,iperm}) where {T,N,perm,iperm}
+function Base.mapreducedim!(f, op::CommutativeOps, B::AbstractArray{T,N}, A::PermutedDimsArray{S,N,perm,iperm}) where {T,S,N,perm,iperm}
+    C = PermutedDimsArray{T,N,iperm,perm,typeof(B)}(B) # make the inverse permutation for the output
+    Base.mapreducedim!(f, op, C, parent(A))
+    B
+end
+function Base.mapreducedim!(f::typeof(identity), op::Union{typeof(Base.mul_prod),typeof(*)}, B::AbstractArray{T,N}, A::PermutedDimsArray{<:Union{Real,Complex},N,perm,iperm}) where {T,N,perm,iperm}
     C = PermutedDimsArray{T,N,iperm,perm,typeof(B)}(B) # make the inverse permutation for the output
     Base.mapreducedim!(f, op, C, parent(A))
     B

base/process.jl

@@ -430,7 +430,7 @@ function open(f::Function, cmds::AbstractCmd, args...; kwargs...)
         rethrow()
     end
     close(P.in)
-    if !eof(P.out)
+    if !(eof(P.out)::Bool)
         waitkill(P)
         throw(_UVError("open(do)", UV_EPIPE))
     end

base/sort.jl

@@ -496,7 +496,8 @@ const SMALL_ALGORITHM  = InsertionSort
 const SMALL_THRESHOLD  = 20
 
 function sort!(v::AbstractVector, lo::Integer, hi::Integer, ::InsertionSortAlg, o::Ordering)
-    @inbounds for i = lo+1:hi
+    lo_plus_1 = (lo + 1)::Integer
+    @inbounds for i = lo_plus_1:hi
         j = i
         x = v[i]
         while j > lo

base/special/exp.jl

@@ -177,7 +177,7 @@ const J_TABLE = (0x0000000000000000, 0xaac00b1afa5abcbe, 0x9b60163da9fb3335, 0xa
 # XXX we want to mark :consistent-cy here so that this function can be concrete-folded,
 # because the effect analysis currently can't prove it in the presence of `@inbounds` or
 # `:boundscheck`, but still the access to `J_TABLE` is really safe here
-@noinline Base.@assume_effects :consistent @inline function table_unpack(ind::Int32)
+Base.@assume_effects :consistent function table_unpack(ind::Int32)
     ind = ind & 255 + 1 # 255 == length(J_TABLE) - 1
     j = @inbounds J_TABLE[ind]
     jU = reinterpret(Float64, JU_CONST | (j&JU_MASK))

base/stream.jl

@@ -409,7 +409,7 @@ function wait_readnb(x::LibuvStream, nb::Int)
         while bytesavailable(x.buffer) < nb
             x.readerror === nothing || throw(x.readerror)
             isopen(x) || break
-            x.status != StatusEOF || break
+            x.status == StatusEOF && break
             x.throttle = max(nb, x.throttle)
             start_reading(x) # ensure we are reading
             iolock_end()
@@ -662,9 +662,11 @@ function uv_readcb(handle::Ptr{Cvoid}, nread::Cssize_t, buf::Ptr{Cvoid})
             elseif nread == UV_EOF # libuv called uv_stop_reading already
                 if stream.status != StatusClosing
                     stream.status = StatusEOF
-                    if stream isa TTY # TODO: || ccall(:uv_is_writable, Cint, (Ptr{Cvoid},), stream.handle) != 0
-                        # stream can still be used either by reseteof # TODO: or write
-                        notify(stream.cond)
+                    notify(stream.cond)
+                    if stream isa TTY
+                        # stream can still be used by reseteof (or possibly write)
+                    elseif !(stream isa PipeEndpoint) && ccall(:uv_is_writable, Cint, (Ptr{Cvoid},), stream.handle) != 0
+                        # stream can still be used by write
                     else
                         # underlying stream is no longer useful: begin finalization
                         ccall(:jl_close_uv, Cvoid, (Ptr{Cvoid},), stream.handle)
@@ -673,6 +675,7 @@ function uv_readcb(handle::Ptr{Cvoid}, nread::Cssize_t, buf::Ptr{Cvoid})
                 end
             else
                 stream.readerror = _UVError("read", nread)
+                notify(stream.cond)
                 # This is a fatal connection error
                 ccall(:jl_close_uv, Cvoid, (Ptr{Cvoid},), stream.handle)
                 stream.status = StatusClosing