Monthly Summary: contrib/anyarray — Generalizing intarray to Any Array Type (May 2026)

Overview

A new contrib module anyarray was proposed to generalize contrib/intarray's set-style operations, boolean query matching, and index support to any array type with a default btree opclass (e.g., bigint[], uuid[], text[]). The patch adds 4,657 lines across 21 files and represents a substantial first submission addressing a long-standing gap in PostgreSQL's extension ecosystem.

No substantive technical review occurred this month — only a procedural note about submission formatting.

Architectural Design

The module is structured into five C source files:

• anyarray.c — Type metadata caching via fn_extra, generic datum comparator
• anyarray_op.c — Set operations: sort, uniq, idx, subarray, intersect, union, difference
• anyarray_bool.c — The anyquery type: parser, evaluator, I/O functions
• anyarray_gin.c — GIN opclasses with per-concrete-type wrappers
• anyarray_gist.c — Polymorphic GiST opclass using signature-based bit vectors

Key Design Decisions

1. Runtime type discovery: Uses lookup_type_cache() with results cached in fn_extra, correctly handling polymorphic calls with varying element types.

2. anyquery type with lazy parsing: Unlike intarray's query_int (integer leaves), anyquery stores text tokens parsed lazily through the element type's input function at evaluation time. This allows the same query value to work against different element types.

3. Polymorphic GiST: Fully type-agnostic signature-based index using dynamically-discovered hash functions. Configurable signature length via reloption.

4. Per-type GIN wrappers: Due to GIN's extractQuery not receiving fn_expr, the @@ strategy requires one opclass per concrete element type (e.g., int8_anyquery_gin_ops, text_anyquery_gin_ops). This is a known limitation requiring GIN AM changes to resolve.

Known Issues and Concerns

• Operator name conflicts: Defines #, &, |, - on anyarray/anyelement — risk of ambiguity with other extensions
• Not marked trusted: Limits adoption in managed PostgreSQL environments
• NULL/multi-dimensional rejection: All operations reject NULLs and multi-dimensional arrays
• GIN scalability: Per-type wrapper approach doesn't scale to arbitrary element types
• GiST picksplit quality: Simple Hamming-weight median split acknowledged as non-optimal
• Stray test artifacts: Accidentally includes regression.diffs and regression.out files

Current Status

The thread received no substantive technical review in May. The only response was a procedural note from David Johnston pointing out that the patch was embedded in the email body rather than attached as a proper patch file. The author needs to resubmit with correct formatting before review can proceed.