SqlTest

SqlTest is the base for the tests that run against a live database. It is abstract, so fant skips it; the concrete subclasses at the bottom of this file bind it to a Dialect.

Each test method gets a fresh instance, a fresh connection, and drops whatever tables it created. No test may depend on another's state.

Postgres array table; created with raw ddl since it never runs elsewhere

Batch tables; see Dialect.batchCols and Dialect.batchAutoCols

Create a table from raw ddl, and register it to be dropped during teardown. For feature-gated tests, where a portable ColDef schema would be pointless because the test only ever runs on one database.

Create the main test table

Connection for the current test method

Insert a row built from a column-name:value map and return whatever Statement.execute returned. Columns are emitted in sorted order so the generated SQL is deterministic.

Insert into the main test table and return the auto-generated key

The single Int column selected by the given sql

A table that is never created

Number of rows in the main test table

A full row of values for the main test table.

Note the float values are exactly representable in binary32, so that v_f32 round-trips through a 4 byte column without widening error, and the times carry no fractional seconds, which mysql datetime truncates. Probing the inexact cases belongs in a dedicated type test.

The row in the main test table with the given key

The one row in the main test table whose v_str is the given key

The v_str values in the main test table, sorted

Postgres array columns map to Fantom Lists; see SqlUtil.fanToSqlObj and SqlUtil.ToFanList.

Whether a batch insert yields keys depends on both the table and the driver, so this pins down all four corners of that.

Binary round-trips through the setBinaryStream path in setParameters

Everything the driver rejects should surface as SqlErr, not as a raw SQLException or a null

An "@" inside quoted text is literal, not a parameter. TokenizerTest covers this at the unit level; here we prove the sql actually runs.

Statement.execute returns one of three shapes depending on the sql

So must the postgres jsonb containment operator; see pod-doc "Escape Sequences". No table is needed, so the sql can be a raw dsl literal.

A parameter missing from the map currently binds as null rather than raising, so a typo in a param name is silent. This pins the current behavior; it is not necessarily the behavior we want.

Exercise the fixture machinery end to end: connect, create, describe, insert, select, and verify both metadata and values.

Round-trip a set of values through every column of the main test table, so that a failure names the exact type and value that broke. This is the conformance matrix: what each database can actually store and hand back unchanged.

A mysql session variable must be escaped inside a prepared statement

Values exercised by testTypeRoundTrip, keyed by logical type. Null is covered separately by testNulls.

Constraints these values respect:

• f32 values are exactly representable in binary32 and carry no more than 6 significant decimal digits. Exactness alone is not enough: mysql float round-trips through roughly 6 significant digits, so 2^24 comes back as 16777200 rather than 16777216
• dec values carry the scale of decimal(9,2). Decimal equality is scale sensitive, since FanDecimal.equals delegates to BigDecimal.equals, so 0d and 0.00d are not equal
• dt and time values are whole seconds, which mysql datetime and time truncate to by default
• char values are exactly 4 chars, because postgres bpchar pads to the declared width while mysql char strips trailing spaces

Compare two cell values. Buf does not define value equality, so binary cells are compared by content.

The main test table; see Dialect.valsCols

Verify every cell in the row is null, apart from the named columns

Verify a row's cells against a column-name:value map. Only the named columns are checked.