DBIO-DB2-0.900004

Changes for version 0.900004 - 2026-11-08

Breaking Changes
- Every identifier is double-quoted as the result class spells it, in DDL, Diff and Deploy alike. Objects an earlier release created in upper case are migrated (export, install, load, or RENAME) before an upgrade.
Storage
- The DSN is built from connection parameters without a DSN (DBIO::AccessBroker::Static, a connect-info hash): dbname alone is the cataloged alias dbi:DB2:<dbname>, dbname with host and port is dbi:DB2:DATABASE=..;HOSTNAME=..;PORT=..;PROTOCOL=TCPIP. host without port, port without host and a missing dbname throw.
- A DBIO::Replicated replicant (HostBound view) over an uncataloged DSN connects to the view's host and port. A cataloged alias and a DSN without HOSTNAME= are refused, naming the form.
- results_exist and its async twins work (FROM SYSIBM.SYSDUMMY1); $rs->rand() orders by RAND(); all six set operations are the ANSI ones core renders.
- $rs->explain runs EXPLAIN PLAN SELECTION FOR and reads the plan back from the EXPLAIN_* tables, found through the new explain_table_schema accessor (SYSTOOLS by default; the name goes through identifier quoting).
- A server warning (SQLSTATE class 01, such as SQL0513W for an unrestricted DELETE or UPDATE) is not an exception; the err, state and errstr stay on the handle. See ADR 0007.
- The character columns of a result are decoded from UTF-8 for a database whose code set is UTF-8. A character LOB (clob, text) is bound as a CLOB.
- An empty string for a blob column and a select or delete comparing a CLOB over 32672 bytes are refused, naming table and column.
- create({}) is INSERT ... (col) VALUES (DEFAULT).
- transactional_ddl is 0; DBIO::DB2::Deploy documents the statement-at-a-time apply and the way back from a partial one.
DDL
- A column without an is_nullable key deploys as NOT NULL; a column that means nullable says is_nullable => 1.
- String defaults are written as escaped literals, and the default of a character column is always a string literal ('00000' on a CHAR(5) stays text).
- unique_constraints are written as named UNIQUE constraints in CREATE TABLE; one over a nullable column is refused (SQL0542N). They are introspected and diffed. See ADR 0009.
- Canonical index declarations (Cake idx, ADR 0061) are written like the db2_indexes entries. An index over columns that the primary key, a unique constraint or an earlier index already covers is left out; an SQL0605W the server sends is a warning.
- boolean is the native BOOLEAN, binary is BINARY(n) up to 255 bytes and VARBINARY(n) up to 32672; beyond that is refused. A text column without a size is a CLOB. The sizeless types varchar, json, varbinary and interval, and the types enum and bit, are refused naming type and column. DECFLOAT is mapped.
- table('schema.table') is the table in that schema; install_ddl writes "schema"."table", the schema is made with its first table, and foreign keys and indexes name it.
- Views are written after the views they read (cycles are refused, naming them); a WITH name and a read qualified with another schema are no views of the set. See ADR 0011.
Introspect
- The columns of the model are the normalized ones of the other drivers (data_type lower case, size, default_value); the catalog form stays internal to the diff. See ADR 0013.
- A foreign key onto a UNIQUE constraint has that constraint's columns as to_columns.
- Every table of the schema is listed, whatever its name; views carry their definition, and the index behind a primary key is in the model with origin 'pk'.
Diff
- Views are created and dropped (DBIO::DB2::Diff::View), a changed definition as drop and create, in the order of what they read.
- An identity column can be added to an existing table and made or unmade identity; what the server refuses is refused naming table and column. See ADR 0012.
- A REORG TABLE follows the column operations that leave a table reorg-pending (DBIO::DB2::Diff::Reorg); see ADR 0008. upgrade reorganises a table already pending before it applies the diff.
- The diff renders the catalog shape: no length on INTEGER or DATE, defaults as stored, identity on create. A default the declaration does not carry is dropped.
- Tables named schema.table are compared in their schema, and the tables of that schema the declaration does not name are never dropped. A foreign key across schemas is refused before the first statement.
- No DROP INDEX or DROP UNIQUE is written for what a dropped column takes along.
Deploy
- diff and upgrade load their classes themselves.
- apply and upgrade are those of DBIO::Deploy::Base: a drop or a lossy column change needs allow_destructive => 1, dry_run => 1 executes nothing. Before the first statement the Db2 checks reorganise pending tables and refuse, naming table and column, an identity column added over rows and a NOT NULL column Db2 has no value for (see DBIO::DB2::Deploy).
- The compare deploy runs entirely in its throwaway schema: tables, indexes and views are qualified, and view bodies resolve through CURRENT SCHEMA. See ADR 0006.
Testing
- DBIO::DB2::Test::Conformance answers the shared case table of core (DDL, Diff and Bind cases), each case in a schema of its own against a real Db2.
Documentation
- The README states the async stance (sync-only; forked and immediate are supported) and lists the query support; ADRs and POD describe the code as it is.
Dependencies
- Requires DBIO 0.900004. Test::Exception is a test prerequisite.
- Every module under lib/ carries a literal $VERSION.