KristofferC

Compiled Pkg apps: open questions

Design questions for compiling Pkg apps to native executables with JuliaC ([apps.<name>] juliac = true). Status of each: Decided (prototype on kc/apps-juliac implements it, with rationale), Open (needs an answer before upstreaming), Deferred (explicitly out of scope for now).

Companion branches: Pkg.jl#kc/apps-juliac, KristofferC/Runic.jl#pkg-apps-juliac (trim=unsafe + preference), KristofferC/ProcessMonitor.jl#pkg-apps-juliac (jtop, trim=unsafe-warn).

KristofferC

PR 780 P1 reproductions

These standalone scripts reproduce the three P1 findings from the review. Run them from the Ferrite.jl repository root. Each script currently fails at the documented invariant; after a fix, it should complete without error.

julia --project=. review_mwes/p1_balanceforest_3d_corner.jl
julia --project=. review_mwes/p1_q2_conformity.jl
julia --project=. review_mwes/p1_subdomain_conformity.jl

KristofferC

#!/usr/bin/env python3
"""Sort a precompile_*.jl file by the `#= NNN.N ms =#` timing comment.

Lines look like:

    #=  465.4 ms =# precompile(Tuple{...})

Some lines are wrapped in ANSI color escapes (e.g. \\033[33m ... \\033[0m) and
end with `# recompile`; those escapes are stripped from all output.

KristofferC

verify_method logs phantom insert_backedges_callee invalidations for CodeInstances that are then promoted (wrong sentinel in the emit guard)

cc @topolarity @Keno — load-time backedge verification / invalidation logging on 1.12 and 1.14-DEV.

TL;DR

During package load, verify_method pushes an insert_backedges_callee entry to the invalidation log for any edge whose per-edge max_valid2 ≠ typemax. But the neutral (unconstrained) value inside verify_method is validation_world, not typemax. So an edge that constrains nothing — e.g. a Core.Binding edge to a

KristofferC

Why Julia's atomic modify doesn't fold to atomicrmw (and when it actually does)

Investigation notes for mwe.jl. Versions tested: 1.12.6, 1.13.0-rc1, 1.14.0-DEV.2212 (062a90bc8c). Source references are to julia master (6e40a4a448, 2026-06-08).

TL;DR

The MWE's code_llvm-based counting is partly misled by a reflection artifact. The real situation, measured on **native code of the actual JIT

KristofferC

#!/usr/bin/env julia
#
#===- git-runic - Runic Git Integration ----------------------*- julia -*--===#
#
# Based on `git-clang-format`, which is part of the LLVM Project, under the
# Apache License v2.0 with LLVM Exceptions (see https://llvm.org/LICENSE.txt).
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
#
#===------------------------------------------------------------------------===#

KristofferC

using SnoopCompileCore

# First load OhMyREPL so its methods are compiled
using OhMyREPL

# Now track invalidations caused by loading SymbolicUtils
invs = @snoop_invalidations using SymbolicUtils;

using SnoopCompile
using SnoopCompile: countchildren

KristofferC

=== SnoopCompile Invalidations Report ===

rebinding Core.Binding(:(Base.HashArrayMappedTries.iterate), #undef, #undef, Any[CodeInstance for MethodInstance for getproperty(::Module, ::Symbol)], 0x00) invalidated:
   mt_backedges: 1: signature Base.HashArrayMappedTries.iterate triggered MethodInstance for getproperty(::Module, ::Symbol) (0 children)


inserting readline(io::VSCodeServer.IJuliaCore.IJuliaStdio) @ VSCodeServer.IJuliaCore ~/dyad1.12/julia-vscode/scripts/packages/IJuliaCore/src/stdio.jl:230 invalidated:
   backedges: 1: superseding readline(s::IO; keep) @ Base io.jl:619 with MethodInstance for readline(::IO) (2 children)

KristofferC

# use with Julia 1.11

using Cassette

Cassette.@context TraceCtx

mutable struct CallTrace
    depth::Int
    pending_newline::Bool
    CallTrace() = new(0, false)

KristofferC

Plots.jl precompile issue

• Create LocalPreferences.toml with

  [PrecompileTools]
  precompile_workloads = false

• Install Plots + HTTP.
• Run julia +1.10 --project --trace-compile=plots_compile.jl
• Run the following: