TurboModula2/SUMMARY.md

Turbo Modula-2 compiler re-read — project summary

Upstream inputs (local only, git-ignored): Reversing-Turbo-Modula2-main/ (disasm + partial decompile) and Turbo-Modula-2-Reloaded-main/ (buildable system). Everything below is versioned: docs/compiler/ (analysis), src/compiler/ (renamed sources).

Problem

The compiler decompiled but was unreadable: every function/procedure/variable a bare number (procN/globalN/paramN/localN).

Done (tags)

• compiler-outline-v1 — docs/compiler/00..70 + TAGGED-SUMMARY.md: folder analysis, per-module architecture, full rename plan.
• compiler-renamed-v1 — src/compiler/COMPILER|SCANNER|ERRORS|CODEGEN.{DEF,MOD}, SYMTAB.DEF, README.md, RENAME-MAP.md: module-level identifiers renamed (e.g. InsertIdent/DeclConst/GetSym/NextChar/ReportError/FlushCodeWindow), cross-module qualified uses updated, 0x1A/0x00 padding stripped.
• compiler-params-v1 — all paramN/localN in the 4 .MOD bodies renamed in-scope (kind/typeDesc/mark/sym/slot/bit/shown/desc/...); fixed a real shadowing bug (FindFileOnDrives: local currentDrive shadowing the global, now startDrive); fixed missed keywordHashFunc and modExtra field. Verified: no numbered identifiers remain in src/compiler/.
• compiler-frontend-map-v1 — src/compiler/FRONTEND.md: every exported head of PASS1 (35), STATEMEN (19), EXPRESS (25+13 Z80 stubs) and SYMTAB bodies pinned (MCode offset + old name + proposed name + evidence), e.g. ~ParseType (02e6), ParseStatement (0566), ParseExpression (1221), Pass1Main (1027). No fake signatures — map only.

Deliberately pending

• Pass1/Statemen/Express bodies (disassembly only, no upstream .MOD).
• SYMTAB.MOD bodies (same).
• GENZ80 native backend.
• Recompile + unassemble.c diff verification of the renamed sources.

Next

Decompile in dependency order: SYMTAB (EnterIdent/LookupChain/ImportAll) → EXPRESS helpers → STMT/STATSE → CASE/FOR → Pass1 types → decls → module heads → SYM writer (docs/compiler/70-rename-plan.md, phase 4).

Verification status (Oct 2026, v1.00 retail toolchain + cpmulator)

Proven by recompile + MCD mnemonic diff (jump targets masked, 00-gaps stripped; dep-slot renumbering benign, proven via slot formula):

• COMPILER.MOD 18/19 byte-identical (linkage names only).
• SCANNER.MOD 34/41 (scanner-verify-01; OOM was FLOAT-of-DIV expr spike, fixed by 1L/2L folding; proc23/24 clean-room, COMPIL VarAddr shim).
• ERRORS.MOD 28/30 (errors-verify-01; heap was import weight — dropped unused Compiler/ComLine; EditPos module + EDITERR overlay load decoded).
• CODEGEN.MOD 34/37 (codegen-verify-01; keywordTable low-byte overlay).
• SYMTAB.MOD prefix 7/14 exact, tail by construction (symtab-wip-04): nesting per call-graph (ee/ec reproduced), record-field views, direct- VAR recursive arms (incl. SRC-vs-entry object fix), REPEAT loops, i * 16 peephole, FindExport hoisted out. ImportAll region PROVEN uncompilable under v1.00-retail (6 experiments; Borland heap ~2x) — full-module verification awaits a bigger-heap Turbo. Genuine fixes banked (TryCompile logic, allocHook, view-typed params, call-graph map). Toolchain rules proven on hardware: DEF-leak, exact DEF/MOD param names, no FORWARD of DEF-declared procs, nominal array checks, transfer-index and transfer-deref bans, zero-cost CHR(ORD())/ORD()/CHR()/= 0, OOM-masking-type-errors. See SESSION.md + RENAME-MAP.md amendments.