Grammar conformance¶

The parsers are generated with ANTLR 4 from combined grammars for SysML v2 and KerML (grammars/SysML.g4, grammars/KerML.g4), taken from hivecore-dev/hcf-runtime and locally patched. This page states exactly what conforms, what was patched, and the one known deviation.

SysML® is a registered trademark of the Object Management Group. This project is not affiliated with or endorsed by OMG, and is not a conformance-certified implementation.

The corpus result: 309 of 309 files parse and build¶

The official SysML-v2-Release repository carries a corpus of 309 .sysml example files. Every file in that corpus parses, and every file builds into the object model, with no lossy fallback. Before grammar patches 6–10, the score was 283 parsed and 280 built.

The corpus itself is not vendored. Two things keep the number honest:

scripts/check_corpus.py reproduces the sweep. It downloads the pinned release commit (the same revision the vendored standard library comes from), then parses and builds every .sysml file in it, printing 309/309 files parse and build. Run it after grammar or builder changes; it is not part of the regular test suite.
Each grammar patch carries regression tests derived from the corpus constructs that exposed it (tests/test_parsing.py, tests/test_builder.py). Those derived snippets – not the full corpus – are what every test run re-checks.

In two places the release BNF and the corpus contradict each other (patches 8 and 10 below). There the grammar follows the corpus, and the README entry for the patch records the conflict.

The patch table¶

Each patch is marked with a LOCAL PATCH comment in the .g4 files. The README carries the full rationale for each; the table below is the summary.

#	Grammar	Patch
1	SysML	`import` accepts an optional visibility prefix, so the spec’s own `import ScalarValues::*;` parses.
2	SysML	Entry transitions accept the spec form `entry; then S;` (upstream required a doubled `then`).
3	both	Unary operators bind tighter than binary operators, so `-3 + 1` parses as `(-3) + 1`.
4	SysML	Metadata and classification use the `@` symbol, keeping the keyword `at` for trigger times (four sites).
5	SysML	Flow ends accept dotted paths, so `flow from a.out to b.in` parses.
6	SysML	State-body transitions put the `then` clause before the action body, per the release BNF, so `accept s : Sig then b;` parses.
7	SysML	`standard` is optional, so a plain `library package P;` parses.
8	SysML	Named send nodes accept the corpus form `action publish send X() via p;` as well as the spec form.
9	both	A `//* ... */` note that closes on its own line lexes as a note instead of swallowing the rest of the line.
10	SysML	Enumerated values accept metadata prefixes, the corpus form `#Security enum secret : Level = 2;`.

The known deviation: operator precedence¶

One deviation from the OMG specification remains, inherited from the upstream grammars:

??, or, and, and implies share one precedence level, and |, &, and xor share another, where the specification separates them;
** is left-associative.

When an expression mixes these operators, parenthesize. The exporter always prints round-trip-safe parentheses, so exported text never depends on the deviation.

Regenerating the parsers¶

The generated parsers are committed under src/longeron/_gen/, so installing and using the package needs no Java. Regenerate only after a .g4 change:

python scripts/generate_parsers.py    # or: pixi run parsers

The script finds Java through JAVA_HOME, PATH, or a conda/mamba environment, and the ANTLR 4.13.2 jar through ANTLR_JAR, ~/.m2, or Maven Central. pixi run parsers needs no setup at all: conda-forge’s antlr package ships the tool and a JDK, and the task caches on its inputs and produces byte-identical output. CI fails if the committed parsers drift from the .g4 sources.