Skip to content

profile-capabilities.md

Latest commit

 

History

History
45 lines (39 loc) · 2.79 KB

profile-capabilities.md

File metadata and controls

45 lines (39 loc) · 2.79 KB
Raw

Dialect Profile Capabilities

See Initial SQL dialect profiles for the SQLite, PostgreSQL, MySQL/MariaDB and independent Dolt starting profiles.

A database dialect profile must publish enough capabilities for an importer to decide whether a dataset can be represented faithfully in one physical wide table.

Required declarations:

  • immutable specification commit, release-candidate/released status, and the published release identifier or NULL;
  • selected profile and database engine separately from transport and driver;
  • raw and normalized server versions, the claimed server-version range, and a separate list of exact CI-tested server versions;
  • CI evidence MUST name exact tested patch versions and the exact container tag or equivalent reproducible service identity; a floating major, minor or latest tag is not exact-version evidence;
  • positive product-identity probes and their normalized results; missing, conflicting, unknown or unclaimed identity must fail before any catalog creation, migration, audit write or dataset mutation;
  • maximum number of physical columns, including __case_ordinal;
  • maximum number of source variables, derived independently from the physical column envelope;
  • maximum identifier length as { value, unit, source, repertoire }, where unit is bytes or characters, plus identifier folding/quoting behavior;
  • reserved-word policy, reserved __ technical prefix, and deterministic physical-name algorithm;
  • exclusive catalog binding mode, namespace or prefix, logical-to-physical relation mapping, and catalog_identity ownership check;
  • binary64-capable numeric type and any exceptional-value constraints;
  • lossless variable-length text type, encoding behavior, and text/value limit;
  • maximum value, row-size and per-statement limits, each with value, unit, source and a basis of theoretical engine limit, exact-version observation, proposed adapter envelope or active effective limit;
  • whether DDL and catalog writes are atomic, or the compensating-cleanup procedure used after a failed write;
  • the exact name of the physical data table and the catalog mapping that associates it with the dataset.

The importer checks these capabilities before object creation. If a source is outside the profile boundary, it fails atomically and records a machine-readable target_capability_exceeded diagnostic. A conforming implementation may support only the subset of SPSS datasets within its declared engine limits.

For MySQL-compatible transports, transport selection is not product identity. In particular, a MySQL URL or driver does not authorize a mysql_mariadb_innodb or dolt engine claim. Each profile's positive identity requirements decide the product. Ambiguity fails closed before any catalog or dataset mutation and leaves zero database mutation.