m-modern-corpus

Snapshot collection of modern (non-VistA) M source from active open-source projects. Used as a validation corpus alongside the VistA corpus (vista-meta/vista/vista-m-host/Packages/) to confirm that m-cli lint rules and tree-sitter-m parsing don't break on idioms outside the VA legacy style.

This project is data-first — its purpose is to be consumed by sibling tooling. The only code is two small helpers under tools/: fetch-corpus.sh reproduces / refreshes the snapshots from upstream, and gen-manifest.py emits the discovery artifacts under dist/. Both read the same provenance table at tools/sources.tsv.

Subdirectories

Path	Source
ewd/	EWD framework
m-web-server/	YottaDB Web Server (M code)
mgsql/	mgsql
ydbocto-aux/	YDBOcto auxiliary routines
ydbtest/	YDB regression test routines

Roughly 4K routines total — the benchmark for the m-cli "default profile = curated daily-lint subset" calibration (3.3 findings/routine).

How it's used

From a sibling checkout of m-cli in the same workspace (~/m-dev-tools/m-cli/):

make lint-modern    # runs `m lint` over this corpus

Results inform rule profile defaults and confirm new lint rules don't false-positive on modern non-VistA M code.

Build / verify

The repo ships two families of make targets. The first regenerates the discovery artifacts — these are declared as the verification_commands in dist/repo.meta.json:

make manifest         # regenerate dist/manifest.json + dist/stats.json
make check-manifest   # drift gate: regenerate, then `git diff --exit-code dist/`

The second family is for snapshot reproducibility — tools/fetch-corpus.sh is the single entrypoint:

make corpus-status    # which subdirs are present vs missing
make corpus-list      # print the upstream provenance table (sources.tsv)
make corpus-fetch     # clone any missing subdirs (no-op when complete)
make corpus-verify    # diff each subdir vs upstream HEAD (drift report, read-only)

For a deliberate snapshot refresh, invoke the script directly:

tools/fetch-corpus.sh refresh <subdir>     # re-clone one
tools/fetch-corpus.sh refresh --all        # re-clone everything

Every fetch/refresh records {upstream HEAD SHA, fetched_at} per subdir in dist/sources.lock.json so the snapshot's provenance is committed alongside the source. The dist/*.json files are deterministic outputs of tools/gen-manifest.py — do not hand-edit them. Python 3, bash 5, and git are the only dependencies.

Status

Maintenance mode. Snapshot only — not auto-synced as upstream projects evolve. Re-sync periodically if the corpus drifts too far behind.

Relationship to other projects

• m-cli — primary consumer; uses this as a validation gate for rules.
• tree-sitter-m — secondary consumer; parses this corpus alongside VistA.
• m-standard — evidence base for which language features are actually used in modern M projects.
• vista-meta — companion corpus (VistA legacy) covering complementary idioms.