[emmanuelbruno]

Why — the H2 v2 uppercase trap

JDBC DatabaseMetaData lookups match the schema/table filters exactly — no case folding. H2 v2 stores unquoted identifiers UPPERCASE; postgres stores them natively lowercase. DBMetadataInspector.inspect passed the %%rdbmsSchema cell filter strings verbatim to md.getColumns / getPrimaryKeys / getIndexInfo / getImportedKeys / getTables (bytecode-verified in the shipped IJava-v1.4.6-pr12.jar; source call sites in this file, reached from JavaDBMSMagics.java:315) → a lowercase filter against an H2 EMPLOYEES table yields empty metadata and the entity renders without attributes/columns — silently. Postgres is unaffected (names natively lowercase, filters typically lowercase too).

ebpro lab workaround so far is content-side only (write the body-line filter uppercase, e.g. EMPLOYEES under PUBLIC) — it works but forces lesson authors to know each store's identifier case policy. This PR obsoletes that.

Change

Case-tolerant filter resolution inside DBMetadataInspector.inspect (the minimal-touch location — it fans out to all six metadata queries, so fixing the caller JavaDBMSMagics:315 alone would leave the other query paths and any future callers exposed):

• Probe the filter as-given first; only when the column lookup returns zero rows, retry UPPER then lower (Locale.ROOT).
• As-given matches — postgres and every correctly-cased call — take the identical code path with identical ordering semantics (single extra getColumns probe only in the currently-broken empty case).
• The TableMetadata display name keeps the as-given value; only query filters are resolved.
• Wildcard filters (e.g. % enumeration) match rows as-given → untouched.

String fSchema = schema;
String fTable = table;
if (table != null && !hasColumns(md, schema, table)) {
    String[][] candidates = {
        {schema == null ? null : schema.toUpperCase(Locale.ROOT), table.toUpperCase(Locale.ROOT)},
        {schema == null ? null : schema.toLowerCase(Locale.ROOT), table.toLowerCase(Locale.ROOT)},
    };
    for (String[] cand : candidates) {
        if (hasColumns(md, cand[0], cand[1])) { fSchema = cand[0]; fTable = cand[1]; break; }
    }
}

The resolved fSchema/fTable replace the verbatim arguments in the five JDBC metadata calls and the three information_schema prepared statements (constraints / check constraints / domains — H2's information_schema likewise stores the uppercase name).

Tests

New DBMetadataInspectorCaseTest (H2 is already on the test classpath — build.gradle: testImplementation 'com.h2database:h2:2.2.220', JUnit 4):

• lowercase filter against UPPER-stored table → columns + PKs resolved, display name preserved;
• as-given match → unchanged columns/order (ID first);
• missing table → still empty (no false resolution).

Local: ./gradlew :test --tests 'io.github.spencerpark.ijava.magics.DBMetadataInspectorCaseTest' green; full :test --tests 'io.github.spencerpark.ijava.magics.*' green (Temurin JDK 25, Gradle 9.7.1).

Base note

Branches from chore/build-correctness (tip = tag v1.4.6-pr12, the lineage that produces the shipped jar). master predates DBMetadataInspector entirely, so it cannot host this change.