@memoarchitect/tools 0.6.7 → 0.8.7

package/README.md

@@ -9,7 +9,7 @@
 </p>
 
 <p align="center">
-  <code>memo-tools 0.6.6</code> &middot; MIT &middot; SysML v2 &middot; ISO 14971 &middot; IEC 62304 &middot; ISO/IEC/IEEE 42010
+  <code>memo-tools 0.8.7</code> &middot; MIT &middot; SysML v2 &middot; ISO 14971 &middot; IEC 62304 &middot; ISO/IEC/IEEE 42010
 </p>
 
 <p align="center">
@@ -69,6 +69,10 @@ memo init my-device --template samd
 memo init gpca --example gpca
 memo ontology show
 
+# Upgrade a pre-native project safely (reports only unless --write is supplied)
+memo convert . --diff
+memo convert . --write
+
 # Validate locally or in CI
 memo validate .
 memo validate . --format junit --output validation.xml
@@ -89,6 +93,12 @@ memo rules coverage
 
 Run `memo --help` or `memo <command> --help` for the complete command surface.
 
+`memo convert` moves legacy project content beneath `model/catalog/`, rewrites
+the corresponding packages/imports/artifact URIs, and creates the native project
+binding entrypoint where needed. It plans before touching the filesystem, refuses
+collisions, and is idempotent. Review its default dry-run output (or `--diff`)
+before using `--write`; then run `memo validate` and your external SysML gate.
+
 ### Jupyter analysis samples
 
 Every `memo init` project includes `analysis/Samples/` with seven portable
@@ -120,16 +130,18 @@ python3 -m pip install "mkdocs>=1.6" "mkdocs-material>=9.5" "pymdown-extensions>
 pnpm run docs:build
 ```
 
-### Compiler and packager selection
+### Toolchain selection
 
-Projects may select external tools in `memo.package.yaml` or
-`memo.config.yaml`. Omit `toolchain` (or use `internal`) to preserve MEMO's
-built-in parser and KPAR writer.
+Each toolchain role is filled by one provider: the **validator** answers "is
+this valid SysML/KerML?", **lowering** answers "what can MEMO ingest from this
+revision?", and the **packager** builds the archive. Omit `toolchain` entirely
+to use MEMO's built-in parser and KPAR writer, which need nothing installed.
 
 ```yaml
 toolchain:
-  compiler: syside       # internal | syside
-  packager: sysand       # internal | sysand
+  validator: syside
+  lowering: internal
+  packager: sysand
   syside:
     executable: ~/.local/bin/syside
     configFile: ./syside.toml
@@ -141,11 +153,24 @@ toolchain:
 ```
 
 `memo validate` and `memo pack` run `syside check --diagnose all
---warnings-as-errors` by default before MEMO's semantic validation, automatically
+--warnings-as-errors` by default when it is the selected validator, automatically
 including the resolved ontology directories. Set `warningsAsErrors: false` or use
-`diagnose: none` to relax the check. `memo pack` delegates
-archive creation to SysAnd when selected. Relative executable and config paths
-resolve from the project directory; bare executable names resolve through `PATH`.
+`diagnose: none` to relax the check. `memo pack` delegates archive creation to
+SysAnd when selected. Relative executable and config paths resolve from the
+project directory; bare executable names resolve through `PATH`. A selected tool
+that is not installed is a clear error, never a silent fallback.
+
+Every leaf under `toolchain` also has a generated flag, so a project can be run
+against a different toolchain without editing its settings:
+
+```bash
+memo toolchain probe                 # which provider, which binary, which version
+memo config effective                # settings + flags + defaults, fully resolved
+memo validate --toolchain.validator syside
+```
+
+`toolchain.compiler` is a deprecated alias kept for existing settings files; see
+`docs/reference/configuration.md`.
 
 ### Core library usage
 

package/corpus/README.md

@@ -0,0 +1,63 @@
+# Conformance corpus
+
+Vendored artifacts from [`Systems-Modeling/SysML-v2-Release`](https://github.com/Systems-Modeling/SysML-v2-Release),
+pinned to one commit and checksummed file by file. `memo conformance` reads
+them; nothing else in the toolchain does.
+
+Upstream paths are preserved verbatim, so every entry in
+`sysml-v2-release/manifest.json` can be checked against the Release repository
+by path and commit without consulting any script here.
+
+## What is here, and what each tree is for
+
+| Tree | Oracle role |
+| --- | --- |
+| `sysml.library/` | Positive parse+link corpus — the normative libraries in textual form |
+| `sysml.library.xmi/` | Declared semantics, before implication |
+| `sysml.library.xmi.implied/` | **Differential oracle** — the reference implementation's own computed output, published as data |
+| `sysml.library.kpar/` | The libraries as published KPARs |
+| `bnf/*.kebnf` | Textual grammar — parser conformance target |
+| `bnf/SysML-graphical-bnf.kgbnf` | Graphical grammar — notation conformance target |
+| `sysml/src/`, `kerml/src/` | Regression corpus — the Release example models |
+
+`manifest.json` records the pinned commit, its date, the library versions read
+off the KPAR filenames, a SHA-256 per file, and one roll-up digest over all of
+them. Every conformance result carries the commit and the digest, so a report
+always says which Release it was taken against.
+
+`baselines/` holds the frozen reports CI regresses against. They live beside the
+corpus rather than with the test fixtures because a baseline is only meaningful
+against one pin — keeping them together means moving the pin puts the stale
+baseline in the same diff.
+
+Neither directory is published: `files` in `package.json` is an allowlist and
+lists neither.
+
+## Commands
+
+```bash
+# re-vendor at the pinned commit
+node scripts/vendor-corpus.mjs
+
+# verify the working tree against the manifest, no network
+node scripts/vendor-corpus.mjs --verify
+
+# run the sweeps and compare against the frozen baselines
+node scripts/check-conformance.mjs
+```
+
+## Moving the pin
+
+Changing the commit changes what conformance means, so it is a deliberate act
+with a reviewable diff:
+
+```bash
+node scripts/vendor-corpus.mjs --commit <sha>
+pnpm build
+node packages/tools/lib/bin/memo.js conformance run --update-baseline
+node packages/tools/lib/bin/memo.js conformance diff-xmi --update-baseline
+```
+
+Also update `PINNED_COMMIT` in `scripts/vendor-corpus.mjs` so a plain re-vendor
+reproduces the new pin. A baseline taken against a different pin is refused
+rather than compared — a corpus bump must never read as a regression.