M2compiler-V3/docs/summary_selfhosting.md

Milestone: V3 self-hosts (byte-identical fixpoint)

Tags v3-fixpoint, v3-bootstrap, v3-selfhosting. Suite 111/111.

What it means

m2compiler-V3 now compiles its own complete source set with itself, reproducing its image byte for byte:

bootstrap/fixpoint.sh
  Stage 1  gm2 -fiso builds V3 (CR -m)             -> m2s1
  Stage 2  m2s1 compiles SymTab QbeGen M2S M2P M2  -> stage2.ssa  -> m2s2
  Stage 3  m2s2 compiles the same set              -> stage3.ssa
  cmp                                               -> FIXPOINT OK
                                                       (1,991,439 bytes)

The compiler no longer needs GNU Modula-2 except as a one-time seed for Stage 1; Stage 2 onward is V3 all the way down.

The pipeline

• Frontend: Coco/R grammar (src/M2.atg → M2S scanner, M2P parser, M2 driver).
• Symbol table (SymTab) + QBE backend (QbeGen) → SSA text.
• qbe → cc (with runtime/syslib/shim.c).
• Two compiler generations coexist: gm2 builds it for the seed; V3 rebuilds it thereafter.

Building blocks delivered along the way

Language: scalars, composites (arrays/records/sets/pointers), control flow, procedures (value/VAR/open/nested + static links/FORWARD), separate compilation (DEFINITION/IMPLEMENTATION, IMPORT, opaque), 64-bit ints, ADDRESS/ADR/SIZE, procedure types + indirect calls, Wirth-style nested modules, short-circuit AND/OR, BITSET, typed set constructors, HALT, and GNU-compatible literals (0x, octal char, LENGTH/ORDL/CAP, &/~).

Runtime/ABI: NUL-terminated CHAR arrays (terminator slot at index = count), inline record array fields, heap/stack descriptor initialisation (nested arrays included), string lib (Strings), FileIO (host for Stage 1; self-hosted runtime/syslib/FileIO for Stage 2+), C shim.

Reproduce

cd /home/eric/Projets/Projets-Modula2/MyWork/m2compiler-V3
./bootstrap/fixpoint.sh                        # FIXPOINT OK
cd compiler && ./build.sh && ./run_tests.sh    # 111/111

Deliberate divergences (per plan)

No generics/interfaces, no ARC/exceptions, QBE backend is terminal (no native codegen), Linux only, and some GNU/classic conveniences accepted for self-hosting. SYSTEM facilities are built-ins, not an importable module.

Where to go next

• Make cross-stage FileIO selection explicit; M2make-driven builds.
• Broaden the stdlib; the Unicode step from docs/plan.md (UCHAR/ UString).
• Optional native backend / additional targets (Blaise parity).