docs: fix terminology, core types, and codec examples

[dimitri-yatsenko]

Summary

Three documentation fixes for DataJoint 2.0 standards:

1. Replace deprecated "antijoin" terminology with "anti-restriction"
2. Replace native int type with core type uint16
3. Fix custom codec examples to use correct storage types and API

Changes

src/tutorials/advanced/sql-comparison.ipynb

• Line 1238: # Antijoin: → # Anti-restriction:
• Line 2586: | Antijoin | → | Anti-restriction |

src/tutorials/advanced/custom-codecs.ipynb

• Line 116: conn_id : int → conn_id : uint16
• Line 256: unit_id : int → unit_id : uint16

src/explanation/custom-codecs.md

• All occurrences: is_external → is_store (8 instances)
• GraphCodec, MedicalImageCodec, FlexibleCodec: return "<hash>" → return "<blob@>"
• BamCodec, LargeDataCodec: return "<object>" → return "<object@>"

src/how-to/create-custom-codec.md

• Example code: return "<hash>" → return "<blob@>"
• Documentation: Updated return value descriptions
• ZarrCodec: return "<object>" → return "<object@>"

Rationale

Terminology (TERMINOLOGY.md)

• Anti-restriction (- operator): Filter rows NOT matching condition
• Restriction (& operator): Filter rows matching condition
• "Antijoin" and "semijoin" are SQL-specific terms, not part of DataJoint's canonical vocabulary

Core Types (Type System specification)

Core DataJoint types (Layer 2) are preferred:

• uint8, uint16, int32, float32 — Standardized, portable, explicit size
• Native types like int, float are discouraged — Backend-specific, lack size metadata

Codec API (DataJoint 2.0)

• Parameter: is_store (not is_external)
• External blob storage: <blob@> (not <hash>)
• Schema-addressed storage: <object@> (not <object>)
• Internal storage: <blob>, bytes, json

Key Fixes

Storage type corrections:

• <hash> was an internal implementation detail, not a public API
• <blob@> is the correct public API for external blob storage
• <object@> is required for schema-addressed storage (plain <object> is invalid)

Parameter name:

• is_store reflects the boolean nature (True = use object store)
• More accurate than is_external which was ambiguous

Verification

• ✅ No other "antijoin" or "semijoin" occurrences
• ✅ Python dataclass SpikeTrain.unit_id: int correctly unchanged (Python type, not DataJoint type)
• ✅ All codec examples now use correct API
• ✅ Storage type documentation updated

Related

• User reports from docs.datajoint.com
• Aligns with TERMINOLOGY.md, Type System specification, and Codec API