Supported Features

pgmt diffs live PostgreSQL catalogs, so it manages what it can read back from pg_catalog. This page lists object coverage and — more importantly — what’s partial, missing, or has known sharp edges.

• ✅ Supported
• 🚧 Partial — works with documented gaps
• ❌ Not supported

Requires PostgreSQL 13 or later.

Object Coverage

Object	Status	Notes
Schemas	✅
Tables	✅	Columns, defaults, identity columns (GENERATED ... AS IDENTITY), type changes, RLS settings
Views	✅	Including security_barrier / security_invoker
Materialized views	❌	Planned
Partitioned tables	❌	Planned
Enum types	✅	Adding values only — PostgreSQL can’t reorder or remove enum values
Composite types	🚧	Create/drop and attribute comments; ALTER TYPE ADD/DROP/ALTER ATTRIBUTE not supported
Domains	🚧	Create/drop; some ALTER DOMAIN constraint operations missing
Range types	✅
Collations	✅	libc, ICU, and builtin (PG17+) providers; deterministic, ICU rules (PG16+); collversion is ignored by design (it varies by machine)
Functions & procedures	✅	Overloading; volatility, STRICT, SECURITY, PARALLEL attributes. OUT/INOUT/VARIADIC parameters and parameter defaults not supported
Aggregates	✅
Operators	✅	All clauses (COMMUTATOR, NEGATOR, RESTRICT, JOIN, HASHES, MERGES), prefix operators
Casts	✅	WITH FUNCTION, WITH INOUT, WITHOUT FUNCTION; see known issues for views using I/O casts
Sequences	🚧	Create/drop, OWNED BY (SERIAL integration); identity-owned sequences are part of their column, not standalone objects; some ALTER SEQUENCE options missing
Indexes	✅	All access methods; partial, expression, and covering indexes; operator classes, collations, storage parameters, tablespaces
Constraints	✅	Primary key, unique, foreign key (actions, deferrable), check, exclusion
Triggers	✅	All timings and events, WHEN conditions, transition tables, column-specific UPDATE OF
Extensions	✅	Extension-owned objects are excluded from management automatically
Comments	✅	On all supported objects, including columns and composite attributes
Grants	✅	Tables, views, schemas, functions, sequences, types; column-level grants; WITH GRANT OPTION; only privilege deltas are emitted. Privileges on objects pgmt doesn’t manage are left alone
RLS policies	✅	All command types, permissive/restrictive, role targeting, USING/WITH CHECK
Roles	❌	By design — see Roles and Permissions

Dependency Tracking

Dependencies come from the PostgreSQL catalogs, not from parsing your SQL: objects are created, altered, and dropped in dependency order, including function signatures, view references, foreign keys, sequence ownership, collations (a domain, table column, composite attribute, index key, or view using a custom collation is ordered after it), and extension-provided types, functions, and operator classes (an index using gin_trgm_ops is ordered after CREATE EXTENSION pg_trgm).

The one gap PostgreSQL itself has: it records no dependencies for the bodies of SQL functions. When a function body references another object, add an explicit -- require: header to the file.

Known Issues and Sharp Edges

• Function body dependencies — PostgreSQL doesn’t track them (including COLLATE clauses inside a body); use -- require: (see above).
• Collation changes are drop + recreate — PostgreSQL can’t alter a collation’s provider, locale, determinism, or rules, so any attribute change recreates the collation and cascades to its dependents. When the cascade reaches a table column, the table recreate is a destructive operation that loses data — flagged for review like any other destructive recreate.
• libc collations name OS locales — the locale must also exist inside the shadow database image, or validation fails with could not create locale. Point shadow at an image (or connection) with matching locales, or prefer ICU collations: their locale strings are self-contained, and ICU is what you want for case-insensitive collations anyway.
• Enum values — append-only; reordering or removing a value requires a manual table rewrite, as in PostgreSQL itself.
• I/O and binary casts in views — casts created WITH INOUT or WITHOUT FUNCTION that are used inside a view or function body aren’t auto-ordered (PostgreSQL records no dependency on them); add -- require: on the cast’s file.
• Extension schema dependencies — extensions created in custom schemas may not order correctly (schema before extension).
• Extension CASCADE drops — dropping an extension with dependent objects requires manual CASCADE handling.
• Complex circular dependencies — some cycles need manual resolution via -- require:.

Out of Scope by Design

• Data migrations — pgmt manages schema structure; data transformations belong in multi-section migrations or external tooling.
• Role management — roles must exist before grants reference them; manage them with a roles file or external tooling.
• Zero-downtime guarantees — depends on the specific change; see Production Operations.

Migration generation, application, validation, baselines, and drift detection are covered in the Migration Workflow and CI/CD guides.