Backports for 1.12.0-rc3 (#59624)

Author: KristofferC

Compiler/src/ssair/passes.jl

@@ -872,14 +872,15 @@ function perform_lifting!(compact::IncrementalCompact,
     return Pair{Any, PhiNest}(stmt_val, PhiNest(visited_philikes, lifted_philikes, lifted_leaves, reverse_mapping, walker_callback))
 end
 
-function lift_apply_args!(compact::IncrementalCompact, idx::Int, stmt::Expr, 𝕃ₒ::AbstractLattice)
-    # Handle _apply_iterate calls: convert arguments to use `Core.svec`. The behavior of Core.svec (with boxing) better matches the ABI of codegen.
+# Handle _apply_iterate calls: convert arguments to use `Core.svec`.
+# The behavior of `Core.svec` (with boxing) better matches the ABI of codegen.
+function lift_apply_args!(compact::IncrementalCompact, idx::Int, stmt::Expr)
     compact[idx] = nothing
-    for i in 4:length(stmt.args)  # Skip iterate function, f, and first iterator
+    for i in 4:length(stmt.args) # Skip `_apply_iterate`, `iterate`, and the function
         arg = stmt.args[i]
-        arg_type = argextype(arg, compact)
-        svec_args = nothing
+        arg_type = widenconst(argextype(arg, compact))
         if isa(arg_type, DataType) && arg_type.name === Tuple.name
+            svec_args = nothing
             if isa(arg, SSAValue)
                 arg_stmt = compact[arg][:stmt]
                 if is_known_call(arg_stmt, Core.tuple, compact)
@@ -900,15 +901,14 @@ function lift_apply_args!(compact::IncrementalCompact, idx::Int, stmt::Expr, 
                     end
                 end
             end
-        end
-        # Create Core.svec call if we have arguments
-        if svec_args !== nothing
-            svec_args[1] = GlobalRef(Core, :svec)
-            new_svec_call = Expr(:call)
-            new_svec_call.args = svec_args
-            inst = compact[SSAValue(idx)]
-            new_svec_ssa = insert_node!(compact, SSAValue(idx), NewInstruction(new_svec_call, SimpleVector, NoCallInfo(), inst[:line], inst[:flag]))
-            stmt.args[i] = new_svec_ssa
+            if svec_args !== nothing
+                svec_args[1] = GlobalRef(Core, :svec)
+                new_svec_call = Expr(:call)
+                new_svec_call.args = svec_args
+                inst = compact[SSAValue(idx)]
+                new_svec_ssa = insert_node!(compact, SSAValue(idx), NewInstruction(new_svec_call, SimpleVector, NoCallInfo(), inst[:line], inst[:flag]))
+                stmt.args[i] = new_svec_ssa
+            end
         end
     end
     compact[idx] = stmt
@@ -1420,7 +1420,7 @@ function sroa_pass!(ir::IRCode, inlining::Union{Nothing,InliningState}=nothing)
                 refine_new_effects!(𝕃ₒ, compact, idx, stmt)
             elseif is_known_call(stmt, Core._apply_iterate, compact)
                 length(stmt.args) >= 4 || continue
-                lift_apply_args!(compact, idx, stmt, 𝕃ₒ)
+                lift_apply_args!(compact, idx, stmt)
             end
             continue
         end

Compiler/test/irpasses.jl

@@ -2045,3 +2045,17 @@ let src = code_typed1(()) do
     end
     @test count(iscall((src, setfield!)), src.code) == 1
 end
+
+# JuliaLang/julia #59548
+# Rewrite `Core._apply_iterate` to use `Core.svec` instead of `tuple` to better match
+# the codegen ABI
+let src = code_typed1((Vector{Any},)) do xs
+        println(stdout, xs...)
+    end
+    @test count(iscall((src, Core.svec)), src.code) == 1
+end
+let src = code_typed1((Vector{Any},)) do xs
+        println(stdout, 1, xs...) # convert tuples represented by `PartialStruct`
+    end
+    @test count(iscall((src, Core.svec)), src.code) == 1
+end

base/client.jl

@@ -267,10 +267,6 @@ function exec_options(opts)
     interactiveinput = (repl || is_interactive::Bool) && isa(stdin, TTY)
     is_interactive::Bool |= interactiveinput
 
-    # load terminfo in for styled printing
-    term_env = get(ENV, "TERM", @static Sys.iswindows() ? "" : "dumb")
-    global current_terminfo = load_terminfo(term_env)
-
     # load ~/.julia/config/startup.jl file
     if startup
         try

base/lock.jl

@@ -767,7 +767,7 @@ end
 
 
 # share a lock/condition, since we just need it briefly, so some contention is okay
-const PerThreadLock = ThreadSynchronizer()
+const PerThreadLock = Threads.SpinLock()
 """
     OncePerThread{T}(init::Function)() -> T
 
@@ -871,7 +871,15 @@ OncePerThread(initializer) = OncePerThread{Base.promote_op(initializer), typeof(
                 state = @atomic :monotonic ss[tid]
                 while state == PerStateConcurrent
                     # lost race, wait for notification this is done running elsewhere
-                    wait(PerThreadLock) # wait for initializer to finish without releasing this thread
+                    # without releasing this thread
+                    unlock(PerThreadLock)
+                    while state == PerStateConcurrent
+                        # spin loop until ready
+                        ss = @atomic :acquire once.ss
+                        state = @atomic :monotonic ss[tid]
+                        GC.safepoint()
+                    end
+                    lock(PerThreadLock)
                     ss = @atomic :monotonic once.ss
                     state = @atomic :monotonic ss[tid]
                 end
@@ -885,7 +893,6 @@ OncePerThread(initializer) = OncePerThread{Base.promote_op(initializer), typeof(
                         lock(PerThreadLock)
                         ss = @atomic :monotonic once.ss
                         @atomic :release ss[tid] = PerStateErrored
-                        notify(PerThreadLock)
                         rethrow()
                     end
                     # store result and notify waiters
@@ -894,7 +901,6 @@ OncePerThread(initializer) = OncePerThread{Base.promote_op(initializer), typeof(
                     @atomic :release xs[tid] = result
                     ss = @atomic :monotonic once.ss
                     @atomic :release ss[tid] = PerStateHasrun
-                    notify(PerThreadLock)
                 elseif state == PerStateErrored
                     error("OncePerThread initializer failed previously")
                 elseif state != PerStateHasrun

base/reduce.jl

@@ -212,7 +212,7 @@ Like [`mapreduce`](@ref), but with guaranteed right associativity, as in [`foldr
 provided, the keyword argument `init` will be used exactly once. In general, it will be
 necessary to provide `init` to work with empty collections.
 """
-mapfoldr(f, op, itr; init=_InitialValue()) = mapfoldr_impl(f, op, init, itr)
+mapfoldr(f::F, op::F2, itr; init=_InitialValue()) where {F,F2} = mapfoldr_impl(f, op, init, itr)
 
 
 """
@@ -231,7 +231,7 @@ julia> foldr(=>, 1:4; init=0)
 1 => (2 => (3 => (4 => 0)))
 ```
 """
-foldr(op, itr; kw...) = mapfoldr(identity, op, itr; kw...)
+foldr(op::F, itr; kw...) where {F} = mapfoldr(identity, op, itr; kw...)
 
 ## reduce & mapreduce
 

base/strings/annotated_io.jl

@@ -200,6 +200,8 @@ function _insert_annotations!(io::AnnotatedIOBuffer, annotations::Vector{RegionA
     end
 end
 
+function printstyled end
+
 # NOTE: This is an interim solution to the invalidations caused
 # by the split styled display implementation. This should be
 # replaced by a more robust solution (such as a consolidation of
@@ -250,6 +252,14 @@ Base.print(io::AnnotatedIOBuffer, s::Union{<:AnnotatedString, SubString{<:Annota
 Base.print(io::AnnotatedIOBuffer, c::AnnotatedChar) =
     (write(io, c); nothing)
 
+styled_print(io::AnnotatedIOBuffer, msg::Any, kwargs::Any) = print(io, msg...)
+
+styled_print_(io::AnnotatedIOBuffer, @nospecialize(msg), @nospecialize(kwargs)) =
+    invoke_in_world(tls_world_age(), styled_print, io, msg, kwargs)::Nothing
+
+Base.printstyled(io::AnnotatedIOBuffer, msg...; kwargs...) =
+    styled_print_(io, msg, kwargs)
+
 # Escape
 
 Base.escape_string(io::IO, s::Union{<:AnnotatedString, SubString{<:AnnotatedString}},

base/terminfo.jl

@@ -303,16 +303,24 @@ end
 """
 The terminfo of the current terminal.
 """
-current_terminfo::TermInfo = TermInfo()
+const current_terminfo = OncePerProcess{TermInfo}() do
+    term_env = get(ENV, "TERM", @static Sys.iswindows() ? "" : "dumb")
+    terminfo = load_terminfo(term_env)
+    # Ensure setaf is set for xterm terminals
+    if !haskey(terminfo, :setaf) && startswith(term_env, "xterm")
+        # For xterm-like terminals without setaf, add a reasonable default
+        terminfo.strings[:setaf] = "\e[3%p1%dm"
+    end
+    return terminfo
+end
 
 # Legacy/TTY methods and the `:color` parameter
 
 if Sys.iswindows()
-    ttyhascolor(term_type = nothing) = true
+    ttyhascolor() = true
 else
-    function ttyhascolor(term_type = get(ENV, "TERM", ""))
-        startswith(term_type, "xterm") ||
-            haskey(current_terminfo, :setaf)
+    function ttyhascolor()
+        haskey(current_terminfo(), :setaf)
     end
 end
 
@@ -352,9 +360,9 @@ Multiple conditions are taken as signifying truecolor support, specifically any
 function ttyhastruecolor()
     # Lasciate ogne speranza, voi ch'intrate
     get(ENV, "COLORTERM", "") ∈ ("truecolor", "24bit") ||
-        get(current_terminfo, :RGB, false) || get(current_terminfo, :Tc, false) ||
-        (haskey(current_terminfo, :setrgbf) && haskey(current_terminfo, :setrgbb)) ||
-        @static if Sys.isunix() get(current_terminfo, :colors, 0) > 256 else false end ||
+        get(current_terminfo(), :RGB, false) || get(current_terminfo(), :Tc, false) ||
+        (haskey(current_terminfo(), :setrgbf) && haskey(current_terminfo(), :setrgbb)) ||
+        @static if Sys.isunix() get(current_terminfo(), :colors, 0) > 256 else false end ||
         (Sys.iswindows() && Sys.windows_version() ≥ v"10.0.14931") || # See <https://devblogs.microsoft.com/commandline/24-bit-color-in-the-windows-console/>
         something(tryparse(Int, get(ENV, "VTE_VERSION", "")), 0) >= 3600 || # Per GNOME bug #685759 <https://bugzilla.gnome.org/show_bug.cgi?id=685759>
         haskey(ENV, "XTERM_VERSION") ||

base/timing.jl

@@ -472,41 +472,93 @@ function gc_bytes()
     b[]
 end
 
-function allocated(f, args::Vararg{Any,N}) where {N}
+@constprop :none function allocated(f, args::Vararg{Any,N}) where {N}
     b0 = Ref{Int64}(0)
     b1 = Ref{Int64}(0)
     Base.gc_bytes(b0)
-    f(args...)
+    @noinline f(args...)
     Base.gc_bytes(b1)
     return b1[] - b0[]
 end
 only(methods(allocated)).called = 0xff
 
-function allocations(f, args::Vararg{Any,N}) where {N}
+@constprop :none function allocations(f, args::Vararg{Any,N}) where {N}
     stats = Base.gc_num()
-    f(args...)
+    @noinline f(args...)
     diff = Base.GC_Diff(Base.gc_num(), stats)
     return Base.gc_alloc_count(diff)
 end
 only(methods(allocations)).called = 0xff
 
 function is_simply_call(@nospecialize ex)
+    is_simple_atom(a) = a isa QuoteNode || a isa Symbol || is_self_quoting(a)
     Meta.isexpr(ex, :call) || return false
     for a in ex.args
-        a isa QuoteNode && continue
-        a isa Symbol && continue
-        Base.is_self_quoting(a) && continue
+        is_simple_atom(a) && continue
+        Meta.isexpr(a, :..., 1) && is_simple_atom(a.args[1]) && continue
         return false
     end
     return true
 end
 
+function _gen_allocation_measurer(ex, fname::Symbol)
+    if isexpr(ex, :call)
+        if !is_simply_call(ex)
+            ex = :((() -> $ex)())
+        end
+        pushfirst!(ex.args, GlobalRef(Base, fname))
+        return quote
+            Experimental.@force_compile
+            $(esc(ex))
+        end
+    elseif fname === :allocated
+        # v1.11-compatible implementation
+        return quote
+            Experimental.@force_compile
+            local b0 = Ref{Int64}(0)
+            local b1 = Ref{Int64}(0)
+            gc_bytes(b0)
+            $(esc(ex))
+            gc_bytes(b1)
+            b1[] - b0[]
+        end
+    else
+        @assert fname === :allocations
+        return quote
+            Experimental.@force_compile
+            # Note this value is unused, but without it `allocated` and `allocations`
+            # are sufficiently different that the compiler can remove allocations here
+            # that it cannot remove there, giving inconsistent numbers.
+            local b1 = Ref{Int64}(0)
+            local stats = Base.gc_num()
+            $(esc(ex))
+            local diff = Base.GC_Diff(Base.gc_num(), stats)
+            gc_bytes(b1)
+            Base.gc_alloc_count(diff)
+        end
+    end
+end
+
 """
     @allocated
 
 A macro to evaluate an expression, discarding the resulting value, instead returning the
 total number of bytes allocated during evaluation of the expression.
 
+If the expression is a function call, an effort is made to measure only allocations from
+the argument expressions and during the function, excluding any overhead from calling it
+and not performing constant propagation with the provided argument values. If you want to
+include those effects, i.e. measuring the call site as well, use the syntax
+`@allocated (()->f(1))()`.
+
+It is recommended to measure function calls with only simple argument expressions, e.g.
+`x = []; @allocated f(x)` instead of `@allocated f([])` to clarify that only `f` is
+being measured.
+
+For more complex expressions, the code is simply run in place and therefore may see
+allocations due to the surrounding context. For example it is possible for
+`@allocated f(1)` and `@allocated x = f(1)` to give different results.
+
 See also [`@allocations`](@ref), [`@time`](@ref), [`@timev`](@ref), [`@timed`](@ref),
 and [`@elapsed`](@ref).
 
@@ -516,11 +568,7 @@ julia> @allocated rand(10^6)
 ```
 """
 macro allocated(ex)
-    if !is_simply_call(ex)
-        ex = :((() -> $ex)())
-    end
-    pushfirst!(ex.args, GlobalRef(Base, :allocated))
-    return esc(ex)
+    _gen_allocation_measurer(ex, :allocated)
 end
 
 """
@@ -541,11 +589,7 @@ julia> @allocations rand(10^6)
     This macro was added in Julia 1.9.
 """
 macro allocations(ex)
-    if !is_simply_call(ex)
-        ex = :((() -> $ex)())
-    end
-    pushfirst!(ex.args, GlobalRef(Base, :allocations))
-    return esc(ex)
+    _gen_allocation_measurer(ex, :allocations)
 end
 
 

deps/checksums/LinearAlgebra-24f5e21cf3a560ca560c5a1759ff21ba68382ebd.tar.gz/md5

@@ -0,0 +1 @@
+93162fc479ba1762028ef917176f45e0

deps/checksums/LinearAlgebra-24f5e21cf3a560ca560c5a1759ff21ba68382ebd.tar.gz/sha512

@@ -0,0 +1 @@
+d6c421048e52d5cf32848cf7a16db1ac269f2c553672b0f1126230a7f5954adb6f2883982efe747bd91fa8135071e1b2f717a12ceb1631c1c7effbf6cece8f4c