remove kwfuncs from Julia

[vtjnash]

This is an internals-only change. These kwsorter methods only exist for dispatch, which means they do not need to be unique. We can optionally have the primary MethodTable contain an extra kwtable::MethodTable field, if this turns out to be slow, since this already introduces the concept of jl_kwmethod_table_for (for better reflection and max_args).

Plus removing this means that we can get wonderful results for invoke now, where previously this would have been broken and non-optimized:

julia> f(x...; kwargs...) = kwargs;

julia> invoke(f, Tuple, stdin=Pipe(), 1)
pairs(::NamedTuple) with 1 entry:
  :stdin => Pipe(RawFD(4294967295) init => RawFD(4294967295) init, 0 bytes waiting)

[aviatesk]

@nanosoldier runbenchmarks("inference", vs=":master")

[aviatesk]

Maybe REPLCompletions also need some tweak? I remember the keyword completion heavily depends on the internal implementations of keyword sorter.

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[vtjnash]

@nanosoldier runbenchmarks(!"scalar", vs=":master")

I deleted that code from REPLCompletions somewhat recently, since it gave almost exclusively wrong answers.

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[aviatesk]

@nanosoldier runbenchmarks("misc" || "sparse" || "tuple", vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[aviatesk]

@nanosoldier runbenchmarks("misc" || "sparse" || "tuple", vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[timholy]

I can confirm that rebasing our "replace dump.c with staticdata.c" branch (teh/make_backref_just_visited) on this one passes the 1.8-era precompile tests.

[vchuravy]

@nanosoldier runtests(ALL, vs = ":master")

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected. A full report can be found here.

[vtjnash]

ExprTools, MethodAnalysis, and ImageView need a closer look, but probably will be okay with this, since everything else seemed fine

[timholy]

2 of the 3 are mine and I'm fine with fixing them. In the case of ImageView, the problematic code is in fact a workaround for #45050. If I understand correctly, it's very likely that this PR fixes that issue. So the failure on ImageView may in fact be the very opposite of a problem 🙂 .

I'd say merge when ready.

[vtjnash]

Okay, it looks like the other package was just using the deprecated parameter here and seems easily fixable also: JuliaTesting/ExprTools.jl#36

The issue #45050 is already apparently fixed, and wouldn't have been affected by this (since it was a lowering issue, and this is a reflection change)

[BioTurboNick]

Related to this PR, I'm seeing kwcall frames show up in stacktraces now.

Will they be displayed differently eventually?

[vtjnash]

I had been seeing that too, and meaning to figure out how to reproduce it. It seems my mistake is here:

diff --git a/base/errorshow.jl b/base/errorshow.jl
index 16190d64e0..65d761aefe 100644
--- a/base/errorshow.jl
+++ b/base/errorshow.jl
@@ -850,10 +850,8 @@ function process_backtrace(t::Vector, limit::Int=typemax(Int); skipC = true)
             code = lkup.linfo
             if code isa MethodInstance
                 def = code.def
-                if def isa Method
-                    if def.name === :kwcall && def.module === Core
-                        continue
-                    end
+                if def isa Method && def.sig <: Tuple{typeof(Core.kwcall),Vararg}
+                    continue
                 end
             elseif !lkup.from_c
                 lkup.func === :kwcall && continue