Targets Summary

Single-page roll-up of every (target ISA x OS) combination Mochi MEP-42 could ship, with status and engineering complexity recommendations.

native-codegentargets

§1 Status legend

  • must-have: Phase 1 ships this combination. Mochi without it is not a credible product.
  • should-have: Phase 1 or early Phase 2. A serious gap if missing.
  • could-have: Phase 2 or later. Valuable to advanced users, not blocking.
  • out-of-scope: not planned for any phase of MEP-42 (might be MEP-43+ or never).

Engineering complexity is a rough 1 to 5 ordinal: 1 = a week or two, 5 = multi-month, multi-engineer effort.

§2 Matrix

Target ISA OS Format ABI File Format File Status Complexity Notes
x86_64 Linux ELF 01_x86_64_sysv 01_elf must-have 2 Cheapest first target. SysV + ELF + static link.
x86_64 macOS Mach-O 01_x86_64_sysv 02_mach_o should-have 3 Universal 2 with arm64 makes this a freebie if arm64 macOS ships.
x86_64 Windows PE/COFF 02_x86_64_windows 03_pe_coff should-have 4 Different ABI; .pdata/.xdata mandatory.
x86_64 FreeBSD ELF 01_x86_64_sysv 01_elf could-have 1 Same as Linux modulo .note.ABI-tag.
x86_64 OpenBSD ELF 01_x86_64_sysv 01_elf could-have 1 Same as FreeBSD; mandatory BTI on capable cores.
x86_64 illumos/Solaris ELF 01_x86_64_sysv 01_elf out-of-scope 1 No reason to skip the work; just no current demand.
aarch64 macOS (Apple Silicon) Mach-O 03_aarch64_aapcs 02_mach_o must-have 3 Apple variadic delta, ad-hoc signing mandatory.
aarch64 Linux ELF 03_aarch64_aapcs 01_elf must-have 2 Pi, Ampere, Graviton, every cloud now has aarch64.
aarch64 iOS / visionOS Mach-O 03_aarch64_aapcs 02_mach_o out-of-scope 5 Provisioning profiles, MH_BUNDLE, App Store machinery.
aarch64 Android ELF 03_aarch64_aapcs 01_elf could-have 3 Bionic loader is stricter; NDK conventions.
aarch64 FreeBSD/OpenBSD ELF 03_aarch64_aapcs 01_elf could-have 1 Pure AAPCS64; trivial increment over Linux.
aarch64 Windows PE/COFF 04_aarch64_windows 03_pe_coff could-have 4 Snapdragon X / X2 Elite; ARM64 xdata is its own bytecode.
riscv64 Linux ELF 05_riscv64 01_elf could-have 3 RVA22/RVA23 hardware exists but user base is small.
riscv64 FreeBSD ELF 05_riscv64 01_elf out-of-scope 2 Niche.
wasm32 (Wasm 3.0 + GC) browser .wasm 06_wasm 04_wasm_module must-have 2 Cheapest backend; massive distribution reach.
wasm32 (Wasm 3.0 + GC) WASI Preview 2 runtime .wasm 06_wasm 04_wasm_module must-have 2 Same emitter, different host imports.
wasm64 runtime .wasm 06_wasm 04_wasm_module could-have 1 Memory64; rarely needed but cheap to support.
ppc64le Linux ELF 07_other_isas 01_elf out-of-scope 4 POWER10 winding down; very small user base.
s390x Linux ELF 07_other_isas 01_elf out-of-scope 5 Big-endian, mainframe-only; no demand.
loongarch64 Linux ELF 07_other_isas 01_elf out-of-scope 4 Chinese market only; ABI well-spec'd but no global demand.
mips* * ELF 07_other_isas 01_elf out-of-scope 3 Sunset ISA; MIPS Inc pivoted to RISC-V in 2021.
Apple GPU macOS Metal AIR 07_other_isas n/a out-of-scope 5 Separate compute pipeline; future MEP territory.
polyglot APE all six APE n/a (post-link) 05_ape_cosmopolitan could-have 3 Distribution feature, Phase 2 bundler tool.

Five host targets, all "must-have" in the table above:

  1. x86_64 Linux (ELF, SysV): cheapest, largest test surface, every CI runner has it.
  2. aarch64 Linux (ELF, AAPCS64): covers Pi, Ampere, AWS Graviton, every modern cloud ARM instance.
  3. aarch64 macOS (Mach-O, Apple ABI): required for Apple Silicon developer adoption.
  4. x86_64 macOS (Mach-O, SysV): comes nearly for free if Mochi ships Universal 2 binaries.
  5. wasm32 with WasmGC: single backend covering browsers and every standalone Wasm runtime (Wasmtime, WAMR, Wasmer, WasmEdge, Spin). Highest leverage per engineering hour.

Estimated Phase 1 effort: ~3-4 engineer-months for first working backends across all five targets, assuming Mochi reuses Go's debug/elf, debug/macho, and writes its own Wasm emitter from scratch.

  1. x86_64 Windows (PE/COFF, MS ABI): required for desktop Windows adoption. Estimated 1-2 engineer-months due to .pdata/.xdata and IAT machinery.
  2. aarch64 Windows (PE/COFF, MS ARM64 ABI): Snapdragon X / X2 Elite laptops, ARM64 dev kits. ~1 engineer-month given the AAPCS64 + PE backends both exist by then.
  3. APE bundler: post-link tool that combines per-OS binaries into a single polyglot APE for distribution. ~2 weeks if cosmocc is the dependency.
  4. riscv64 Linux (ELF, RV64GC LP64D): RVA22/RVA23 hardware is real. ~1 engineer-month.

§5 Explicitly out-of-scope for MEP-42

  • iOS / iPadOS / visionOS / tvOS / watchOS: requires Apple developer credentials, provisioning profiles, App Store review. Defer to a dedicated MEP for mobile.
  • ppc64le, s390x, loongarch64: real but small user bases. Add on demand.
  • MIPS: sunset.
  • GPU compute (Metal AIR, CUDA PTX, ROCm, SPIR-V, WGSL): a separate MEP for SIMT codegen.
  • macOS x86_64 once Apple removes Rosetta: handle when it happens (likely 2027+).
  • ARM64EC: deferred until Mochi has a story for x64 plugin interop on Windows.

§6 Critical decisions MEP-42 must make

  1. Which native object writers does Mochi vendor (write itself) vs invoke (shell out to lld, ld, link)? Recommend vendor minimal ELF + Mach-O + PE writers, invoke system linker.
  2. Static vs dynamic linking default? Recommend static for Phase 1 (simpler, fewer runtime dependencies); dynamic in Phase 2.
  3. Linker choice: bundle lld, depend on system linker, or both? Recommend system linker with lld as fallback when missing.
  4. Debug info: DWARF on Linux/Mach-O/Wasm; CodeView (Phase 1) then PDB (Phase 2) on Windows.
  5. Cross-compilation: officially supported in Phase 1 for the five must-have targets from any host? Recommend yes; Go's standard cross-compile facilities make this cheap.