razeri-plugin-api
Extension traits and registries for building Razeri plugins.
This is the single crate third-party authors depend on to extend Razeri with new database dialects, storage backends, LLM providers, AI model adapters, language renderers, and event sinks.
Extension Points
| Trait | Registry | Purpose |
|---|---|---|
PlatformDialect |
PlatformRegistry |
SQL dialect for a new database |
StorageProviderFactory |
StorageRegistry |
Storage backend (S3, Azure, etc.) |
LlmClientFactory |
LlmRegistry |
LLM provider (Ollama, OpenAI, etc.) |
LlmModelAdapter |
ModelRegistry |
Model capabilities and routing |
LanguageRendererFactory |
LanguageRegistry |
Code generation target language |
EventSink |
EventRegistry |
Structured telemetry sink |
The rendering traits themselves live in
razeri-plugin-contract,
which this crate depends on and re-exports as razeri_plugin_api::razeri_plugin_contract.
Quick Start
use Arc;
use EngineVersion;
use ;
use ;
;
// SqlRenderer: 4 required methods; everything else defaults to ANSI behaviour.
// Remaining traits: all methods have defaults — each override is a decision
// your engine states about itself.
// One line makes the crate a plugin, loaded from a directory or linked in.
export_plugin!;
Running a plugin
A plugin reaches razeri one of two ways, and export_plugin! serves both:
- Loaded from a plugin directory. Build the crate as a
cdyliband put the library in the directoryrazeri --plugin-dirnames. It shares Rust types with the razeri that loads it, so it must be built with the same compiler, against the same versions of this crate andrazeri-plugin-contract, from the same source.razeri plugin listprints what that razeri was built with, and a library built any other way is refused with a message saying what to rebuild it with. - Linked into a razeri you build. Depend on the crate from a program whose
mainruns razeri's command line, and the dialect is one of the platforms it renders. One build means nothing to match.
The book's Dynamic loading and Linking a dialect pages give both in full. The book comes with each release, on the releases page.
Proving a plugin
The conformance kit runs from your own test suite and grades the plugin the way Razeri grades its shipped dialects — defaulted decisions, contradictory capability answers, missing refusal wording:
let report = check;
assert!;
Your dialect's names are yours: name is the one Razeri reports, aliases is
every other spelling a design may use for the same platform, and the resolver
reads both off the dialects that are loaded. A clean report never depends on
anything inside Razeri — only on not claiming a spelling another loaded platform
already answers to.
Stability
This crate and razeri-plugin-contract follow semantic versioning while
the major version is zero. A trait gains a method only with a default body,
so a new method costs an existing plugin no line of code. A change that does
break a plugin's build comes only in a minor release, and the
CHANGELOG.md in razeri-plugin-contract's package says what to change. A method that leaves does so as #[deprecated] for one minor
version before it is removed, so a plugin sees the warning before the
removal, not after.
Features
dynamic(default) — Enables runtime plugin loading via shared libraries (libloading).
Documentation
- The book's Plugin Authoring Guide and Plugin Architecture pages, in the book that comes with each release on the releases page.
- API reference, on docs.rs.
License
Dual-licensed under MIT or Apache 2.0, at your option — this crate is the open surface of an otherwise proprietary workspace, so plugin authors depend on it freely. The rest of Razeri is covered by the licence it is released under (see its §4, "Plugins are open").
Security
Report a vulnerability as Razeri's security policy says.