Session handoff — 2026-10-08

Branch generics-inc1 (off ast-stage-c). Working tree clean.

Current state: suite 189/189, FIXPOINT OK (3,490,054 bytes).

Where the branches/tags stand

ast-stage-c — the L7 refactor + language work (tags below), HEAD 0f611c0 docs: GOTO/labels landed.
generics-inc1 — ast-stage-c + Increment 1 of generics, HEAD 55d2e7c docs: generics inc1 status + inc2 plan.

Tags on ast-stage-c (chronological): v3-lower7-flip, v3-lower7-decoupled, v3-lower7-complete, v3-fix-arrrecfield, v3-arrrecfield-complete, v3-status-2026-10-08, v3-composite-results, v3-chain-calls, v3-goto.

What this session did

L7 refactor — done. Lower is the only emitter; the inline emit was deleted from statements, control flow, declarations, expressions/designators, calls and ProcHeading. Found and fixed a latent VAR-actual note-ring bug (RestoreSession now clears the ring). Swept dead code (0 unreferenced procs). Retired -lower and the degenerate byte-compare tests (suite 204 → 181).

Language / correctness work (all on ast-stage-c):

array-of-record inline length words — small arrays expand inline, large ones get a generated run-time init loop (t_arrrecfld, t_arrrecbig).
statement-context result suffixes (F()^.x := e, t_stmtsfx).
composite function results (record/array/set, sret) — which also fixed methods with local variables (their block decls were never stored); t_retrec, t_retmeth.
class results + chained post-call method calls (F().M(), t_chain).
GOTO/labels — required splitting Design into Design + DesignTail to resolve the LL(1) label/designator clash; t_goto, t_gotobad.

Docs refreshed: STATUS.md, analysis_gaps.md, features.md, analysis_generics.md.

Generics (the active thread)

Syntax chosen: ISO 10514-2 generic modules + refinement (../Generics/m2generic-syntaxe-analysis.md).

Increment 1 — done (e1d0472). Grammar recognition of the four module categories: GENERIC DEFINITION/IMPLEMENTATION MODULE M (E : TYPE); … and DEFINITION/IMPLEMENTATION MODULE X = M(actuals); …. Formals are unresolved alias types (SymTab.EnterTypeParam); refinements are recorded. A refinement still yields an empty module.

Increment 2 — the next step. Instantiate by re-parsing the generic source with a substitution scope, driven by a driver session pre-pass:

record each generic's def/impl file path when parsed;
on a refinement, re-open the generic, bind formals to actuals, name the module X, dedup, and lower it before the importer;
move the normal body's BeginDef into the ; branch so the refinement does not pre-register X; add SymTab.EnterTypeParamBound; add a grammar instantiation context (instActive/instName/instActual) + accessors (IsGenericUnit, IsRefinementUnit, GetRefinement, SetInstantiation); add the driver re-parse logic.

Details and the full plan are in docs/analysis_generics.md (top note).

Also still open (from `docs/STATUS.md` / `analysis_gaps.md`)

L7 item 7 (delete SaveSession/RestoreSession/SetNoEmit) — blocked on the grammar's last suppressed writers; low payoff.
Library holes (device/IO framework, docs/wip/chanconsts.*, streamfile.*).
Robustness: M2 exits 0 on errors / aborts on a missing input file; weak error recovery; fixed table limits.
The latent size-sensitive V3 miscompile (removing ASTDUMP from M2.atg breaks the fixpoint).
Generics increments 3–4 (multiple/const params, proc params, constraints, GENERIC CLASS).

How to resume

cd compiler && ./build.sh && ./run_tests.sh   # 189/189
cd .. && bootstrap/fixpoint.sh                # FIXPOINT OK
git log --oneline -5

session-handoff-2026-10-08.md 3.8 KB История Исходник