openehr-oracle

openEHR persistence for Oracle Database 23ai — the schema dialect.

Conformance level: Dialect#

This crate emits DDL for the shared openEHR schema. It does not contain a store. There is no driver dependency, no connection handling, no implementation of Store, and no statement in this crate has ever been submitted to a Oracle Database 23ai server.

See openehr-store/spec/conformance.md for what each level means and why they are stated this bluntly.

use openehr_oracle::OracleDialect;
use openehr_store::ddl_script;

println!("{}", ddl_script(&OracleDialect));

What this crate owns#

Four things: type spellings, identifier quoting, placeholder style, and how the engine enforces append-only. Everything else — which tables exist, which columns, which indexes, the projection from openEHR objects onto rows, the commit rules, the conformance suite — lives in openehr-store and is shared by all five engines.

That boundary is deliberate, and this crate is the reason it exists. The sibling FHIR monorepo in this repository gave each of six ports a full copy of the DDL generator, and its Oracle port spent the fork's whole life emitting MySQL types (F-08) — a file copied with three lines changed, which nothing ever compared. A dialect that owns only spellings cannot do that; openehr-sqlite/tests/dialects.rs compares all five; and this crate's own tests assert that its DDL contains no TINYINT.

Oracle Database 23ai specifics#

Decision	Why
`VARCHAR2(n CHAR)`, with `CHAR` semantics	`VARCHAR2(255)` means 255 bytes on Oracle, so a name in a non-Latin script would be rejected at about a third of its length.
`CLOB` for JSON and unbounded text	`VARCHAR2` maxes out at 4000 bytes; anything unbounded must be a LOB.
`NUMBER(19)` for integers, `NUMBER(1)` for booleans	Oracle SQL has neither an integer nor a boolean type, whatever PL/SQL offers. `NUMBER(19)` is the range of the `i64` the row types use.
Quoted, case-sensitive identifiers	Unquoted Oracle identifiers fold to upper case, which would make this schema's lower-case names disagree with every other engine's.
No `IF NOT EXISTS`	Oracle has none before 23ai, and this script is meant to run on 19c too.
An identifier-length test	Oracle is the only one of the five with a limit short enough to hit, and an over-long name fails at `CREATE TABLE` and nowhere else — so a schema change made against another engine would break here alone.

Every instant is stored twice, and that is the point#

openEHR times are ISO 8601 strings with deliberate partial precision: 2024-05 is a date known to the month, and it is not 2024-05-01. A native timestamp column silently completes it — fabricating a clinical fact — and normalises the lexical form, breaking round-trip fidelity.

So each time occupies …_text (authoritative, exact) and …_utc (derived, nullable, for ordering). The derived column is NULL whenever the instant is not established, which is the same answer the library gives, so SQL and Rust cannot disagree about one record.

Testing#

cargo test

The tests are golden: they assert the SQL this crate emits, including assertions that it is not another engine's SQL.

Licence#

MIT OR Apache-2.0.