Published OData services could not build from MDL; JAR dependencies never resolved; init/settings/grammar fixes (formula1 findings)

[ako]

Works through the findings from building an F1 data app: a DuckDB-backed, non-persistable entity published over OData v4. Everything was reproduced against real MxBuild before being fixed, and re-verified after (mx check → 0 errors on Mendix 11.12.1 — 11.13.0 was not cached here, so version numbers below are 11.12.1 unless stated).

Two of the reported items did not survive checking; those are called out rather than quietly "fixed".

The headline: every published OData service from MDL was unbuildable

Two defaults CREATE ODATA SERVICE never set, each fatal on its own, neither visible to mxcli check:

CE0729 "The service name should not be empty". Name (the document) and ServiceName (the name in the OData metadata document) are different properties, and CREATE set only the first. The consumed path had defaulted the same field to the document name for CE0339 all along; the published path never got the line. An explicit ServiceName: still wins, and create or modify heals an empty one.

CE7375 "Attribute ID … must be published and be the key when associations are exposed as an associated object id" — firing with no associations exposed at all.

The finding framed this as a non-persistable-entity problem. It is not:

entity	key	PublishAssociations	mx check
persistent	unique attribute	false (the old default)	CE7375
persistent	unique attribute	true	0 errors
non-persistable	attribute	false	CE7375
non-persistable	attribute	true	0 errors

false means "associations as an associated object id", which Mendix only allows when the system ID is the published key — while MDL's expose (Attr (KEY)) publishes an ordinary attribute. So the default broke every published service; non-persistable is merely where no workaround exists.

Defaulting to true is not a preference, it is the only value that can build from the MDL people write. Tri-state so an explicit false survives, and create or modify no longer flips a stored value the script never mentioned.

Nothing in-repo caught either default: doctype-tests/10-odata-examples.mdl sets both properties explicitly.

The other silent one: a declared JAR dependency never reached the classpath

ALTER MODULE X ADD JAR DEPENDENCY (…) wrote the coordinate, list jar dependencies reported it, the build went green — and the runtime threw No JDBC driver found in app for URL.

The write was never wrong. Declaring and resolving are separate steps: mx sync-java-dependencies <project.mpr> downloads into vendorlib/, Studio Pro runs it when you edit Module Settings, and nothing ran it headless. Measured on 11.12.1: a full mxbuild --target=deploy emits a build.gradle with no dependencies block and downloads nothing; the sync command then fetches the jar.

That answers the open question in the report — neither "mxbuild skips Maven resolution" nor "mxcli writes where MxBuild cannot read". (Found by strings mx.dll | grep -i dependenc surfacing ISyncJavaDependenciesRunner, then mx --help.)

Wired at three levels so it cannot stay silent:

• the executor says so the moment it writes a coordinate that is not in vendorlib/;
• mxcli sync-java-deps does the resolution on demand, with --check exiting non-zero as a build gate;
• run --local vendors anything missing before boot, so the warm loop works from a fresh clone.

Resolution needs network, so each call site is best-effort and names the command to retry. Falling back to another Mendix version's mx is now stated rather than silent — a mismatch otherwise surfaces as an mpr-format complaint that reads like a corrupt project.

Corrections to the report

• PublishAssociations is accepted on CREATE today (visitor_odata.go — the grammar takes a generic property list). A one-shot script could always express it; what was missing was documentation. The real defect was the default.
• Pagesize: is not silently dropped. The visitor lowercases before matching, so casing is never a typo. A test pins that so the new rule cannot start flagging it.

Check-time gaps the build does not cover

Silently discarded OData properties (MDL-ODATA01). The grammar accepts any name: value pair and the visitor's switch had no default, so ReadMicroflow: or ServiceNam: parsed, checked, executed "successfully" — and left the model without the property. The name is lost in the visitor, so that is where it is recorded; a check-time validator reports it before anything is written, guessing the intended property rather than only listing the known ones.

Unreal database-connection types (MDL-DB01). The skill's table listed Redshift and SQLServer; neither is in Studio Pro's picker, read out of the shipped bundle and identical on 11.10.0, 11.12.1 and 11.13.0: MSSQL, MySQL, Oracle, PostgreSQL, Snowflake, BYOD. mxcli writes the string through and mxbuild does not validate it either — type 'Redshift' builds 0 errors and simply does not connect. A warning rather than an error: the set is version-specific.

BYOD ("Other") was the missing entry that mattered — it skips the driver-presence check, which is what makes an unsupported JDBC driver usable. Now confirmed working against a live runtime, so the docs were the only thing in the way.

Query options and the DESCRIBE round trip

Countable / SkipSupported / TopSupported were literal true in the BSON writer with no MDL above them. Countable is not cosmetic: it forces every read-microflow-backed resource to declare a System.ODataResponse parameter and compute a count. Now publish-entity properties, tri-state because they default to true. Verified: Countable: No with a read microflow that takes no parameters builds 0 errors.

DESCRIBE emitted MDL it could not parse. Three slips, plus a fourth not in the report: the stored CallMicroflow: spelling, the entity type's exposed name where the entity set's belongs (silently renaming the set on a re-exec), and fully-qualified member names in an expose (...) clause that takes bare ones. Proved by round trip: describe → check parses → drop → exec the output → describe again → byte-identical, and mxbuild reports 0 errors on the rebuilt model.

MDL papercuts

$Total = 5; now parses. It failed with "no viable alternative at input '$Total=5'" while DECLARE $Total Integer = 0; worked — and so did $X = HEAD($List), $X = create M.E (…) and $X = execute database query …. Assignment existed only as a prefix on activity statements plus a SET $Var = … statement, so a plain value needed a keyword nothing suggested. SET is now optional; both spellings produce the same AST.

Regressions were checked with a control binary rather than by reading the grammar: stash the .g4, regenerate, build bin/mxcli-control, sweep every script in mdl-examples with both. Thirteen fail — the same thirteen, all pre-existing.

mxcli test tests/ -p app/App.mpr no longer fails for a tests/ next to the .mpr. The fallback applies only when the CWD-relative path does not exist, so a tests/ in both places still resolves to the one you are standing in.

Smaller ones

• mxcli init stopped inventing a project. From a solution root it reported success and wrote tooling pointing at project.mpr, a file that does not exist. It now looks one level down and lets the candidate count decide: none → warn; one → name it and initialise that directory; two or more → refuse and print the command naming a specific app.
• describe settings configuration '<name>' parses (the read form of an existing write statement), and show settings configurations carries ApplicationRootUrl.
• The ANTLR pin lives where the failure happens — the Makefile holds it, the error quotes it, make -C mdl/grammar bootstrap does the install, and the README says the build needs network and a JVM, not only Go.
• mxcli new step 6 says "links" — on Linux it hard-links the binary you ran rather than downloading the nightly.
• Documented ReadMode: microflow …, ServiceName and the query options in mxcli syntax odata publish, with a worked non-persistable example. The skill's own example carried PublishAssociations: No, which cannot build; removed with the reason. Added the two dynamic query rules (every declared parameter still needs a value; a {param} placeholder concatenates into a path).

Not done

• The solution-level bootstrap hook is a feature, not a fix; this adds only the sibling-project discovery it would need.
• Ergast being dead is a dataset fact, not an mxcli defect.
• The Gradle daemon surviving pkill is correct behaviour — it is a daemon, reused across builds.

Verification

• Every fix has a unit test; the mxbuild-observable ones were A/B'd against a pre-fix binary.
• Three MDL scripts under mdl-examples/bug-tests/ execute against a real .mpr and build 0 errors with no workarounds.
• MDL-ODATA01 and MDL-DB01 checked against every script in mdl-examples for false positives: none.
• The JAR-dependency path verified end to end: declare → run --local --setup → jar in vendorlib/.
• Rebased onto current main after mxcli-todo findings: nine fixes from bootstrapping an app (check-parity, first-build cleanliness, theming, diagnostics) #107/Fix silent drops of conditional expressions with keyword-named functions #108/Add test endpoint for warm test re-runs without restart #109; grammar regenerated, go build ./... clean, full go test ./... green.
• Symptom rows appended to .claude/skills/fix-issue.md for each fix.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JXnEgoc2NQP1Y2TWMCMXC4