# topspeed-grammar

Corpus harness for the historical **TopSpeed / JPI V3** dialect, using
the sidecar grammar `compiler/src/TopSpeed-V3-M2.atg`.

The sidecar is a *grammar* deliverable: it parses the TS-V3 dialect
(pragmas, `H`/`C`/`B` literals, `::=`, anchored subranges, address
constructors, `CLASS`/`IS`/`VIRTUAL`, manifest constants, tagless and
trailing-`|` variant records, ...). Constructs the current backend
cannot lower yet still **parse** and end with the standard
`SemError(230)` marker (parse-now / lower-later), exactly like
`M2.atg`.

## Build

```sh
./build.sh        # -> build/TSM2   (build/ is git-ignored)
```

It copies the grammar, the frames and the V3
`SymTab`/`QbeGen`/`FileIO` sources into `build/`, runs `CR -m -C`,
compiles with `gm2 -fiso` and links `TSM2`. Needs the project `CR`
(`$HOME/Projets/Projets-Modula2/MyWork/CocoGm2/CR`) and GNU Modula-2.

The sidecar uses **its own scanner frame** — `scanner.frm` in this
directory — which adds TopSpeed **conditional compilation**
`(*%T cond*)` / `(*%F cond*)` / `(*%E*)`: a `CondFilter` blanks the
inactive regions of the source buffer (CR/LF kept) before scanning.
The define set is `{_fdata, _mthread, _fptr, _fcall}` (see `InitDefs`);
tune it there.  V3's own `compiler/src/scanner.frm` is left untouched,
so the sidecar never affects the self-hosting compiler.

## Corpus check

```sh
./corpus.sh [corpus-dir]     # default: $HOME/bin/Dos/M2/TS-V3
```

Runs `TSM2` over every `.DEF`/`.MOD`, copying each into a scratch tree
so the `.LST` listings never touch the corpus, and reports:

- **parse-OK** — only the deliberate `230`/semantic markers,
- **syntax-bad** — a real Coco/R error (`'x' expected`, `invalid X`).

Current result on the TS-V3 tree (662 files, 352 `.DEF` + 310 `.MOD`):
**644 parse-OK / 18 syntax-bad** — see
`docs/summary_topspeed-corpus.md`.