stemedb/.claude/guides/backend/api-endpoints.md
jml 4012791e7e fix(api): enable non-strict mode for URL-encoded bracket notation
## Problem
Dashboard sends URL-encoded query parameters:
  ?sources%5B%5D=rfc&sources%5B%5D=owasp
  (%5B = '[', %5D = ']')

But QsQuery extractor used strict mode, which rejects encoded brackets:
  Error: "Invalid field contains an encoded bracket"

Result: All corpus filters in the dashboard failed silently.

## Solution
Changed QsQuery to use serde_qs non-strict mode:
  Config::new(5, false) // false = non-strict

Now accepts BOTH:
  - Literal brackets: ?sources[]=rfc
  - Encoded brackets: ?sources%5B%5D=rfc (browsers)

## Verification
 URL-encoded query: ?sources%5B%5D=rfc&sources%5B%5D=community
   Returns: 24 items (was: error)
   Logs: sources=Some(["rfc", "community"]) 

 Literal brackets: ?sources[]=rfc (still works)
 All 4 extractor tests pass (added encoded brackets test)
 Clippy clean (0 warnings)

## Files Changed
- crates/stemedb-api/src/extractors.rs: Use non-strict Config
- crates/stemedb-api/README.md: Document QsQuery usage
- .claude/guides/backend/api-endpoints.md: Add best practices
- CLAUDE.md: Reference extractors documentation

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-02-09 16:11:25 +00:00

222 lines
6.2 KiB
Markdown

# Add an API Endpoint
**When to use:** You need to add a new HTTP endpoint to the Episteme API.
## Prerequisites
- Familiarity with `axum` handler patterns
- Understanding of `utoipa` derive macros
- Read [ai-lookup/services/api.md](../../../ai-lookup/services/api.md) for architecture overview
## Quick Start
```bash
# After adding endpoint code:
cargo build -p stemedb-api
cargo test -p stemedb-api
cargo clippy -p stemedb-api -- -D warnings
```
## Step-by-Step
### 1. Define DTO Types
Create request/response types in `crates/stemedb-api/src/dto.rs`. These derive BOTH serde (for JSON) and utoipa (for OpenAPI docs):
```rust
#[derive(Serialize, Deserialize, ToSchema)]
pub struct MyRequest {
/// Description appears in OpenAPI docs.
#[schema(example = "Tesla_Inc")]
pub subject: String,
}
#[derive(Serialize, Deserialize, ToSchema)]
pub struct MyResponse {
pub hash: String,
pub status: String,
}
```
Rules:
- DTOs live in `dto.rs`, NOT in handler files
- Implement `From<stemedb_core::types::X>` for conversions to/from internal types
- Use `#[schema(example = "...")]` for meaningful OpenAPI examples
- Never expose internal rkyv types in the public API
### 2. Write the Handler
Create handler in `crates/stemedb-api/src/handlers/{name}.rs`:
```rust
#[utoipa::path(
post,
path = "/my-endpoint",
request_body = MyRequest,
responses(
(status = 202, description = "Accepted", body = MyResponse),
(status = 400, description = "Validation failed", body = ApiError),
),
tag = "assertions"
)]
pub async fn my_handler(
State(state): State<AppState>,
Json(req): Json<MyRequest>,
) -> Result<(StatusCode, Json<MyResponse>), ApiError> {
// Convert DTO -> internal type
// Call library code (QueryEngine, Ingestor, etc.)
// Convert result -> DTO
// Return
}
```
Rules:
- Handlers are thin: convert types, delegate to library code, convert back
- No business logic in handlers
- Always return `Result<_, ApiError>` for consistent error responses
- Use `tag = "..."` to group related endpoints in docs
### 2.1. Query Parameters with Arrays
**CRITICAL:** If your query parameter DTO contains `Vec<T>` or `Option<Vec<T>>`, you MUST use `QsQuery` instead of standard `Query`.
#### Use `QsQuery` for Array Parameters ✅
```rust
use crate::extractors::QsQuery;
#[derive(Deserialize, ToSchema)]
pub struct FilterRequest {
/// List of sources to filter by
pub sources: Option<Vec<String>>, // ← Array type
pub limit: usize,
}
#[utoipa::path(
get,
path = "/my-endpoint",
params(
("sources" = Option<Vec<String>>, Query, description = "Filter sources"),
("limit" = usize, Query, description = "Max results"),
),
// ...
)]
pub async fn my_handler(
State(state): State<AppState>,
QsQuery(params): QsQuery<FilterRequest>, // ✅ CORRECT
) -> Result<Json<MyResponse>> {
// Dashboard sends: ?sources[]=rfc&sources[]=community&limit=10
// params.sources = Some(vec!["rfc", "community"])
}
```
#### Never Use Standard `Query` with Arrays ❌
```rust
// ❌ WRONG - Will silently fail!
pub async fn broken_handler(
State(state): State<AppState>,
Query(params): Query<FilterRequest>, // sources will always be None!
) -> Result<Json<MyResponse>> {
// Dashboard sends: ?sources[]=rfc
// Result: params.sources = None (filter doesn't work!)
}
```
#### Use Standard `Query` for Scalar Parameters Only ✅
```rust
use axum::extract::Query;
#[derive(Deserialize, ToSchema)]
pub struct SimpleRequest {
pub limit: usize,
pub offset: usize,
pub category: Option<String>, // All scalars, no arrays
}
pub async fn simple_handler(
State(state): State<AppState>,
Query(params): Query<SimpleRequest>, // ✅ CORRECT
) -> Result<Json<MyResponse>> {
// ?limit=10&offset=0&category=security
}
```
**Why:** The StemeDB Dashboard uses JavaScript's `URLSearchParams` which generates bracket notation for arrays (`?filters[]=a&filters[]=b`). Standard `axum::extract::Query` uses `serde_urlencoded` which doesn't support brackets. `QsQuery` uses `serde_qs` which does.
**Quick Check:**
- DTO has `Vec` or `Option<Vec>` field? → Use `QsQuery`
- All fields are scalars? → Use `Query` or `AxumQuery`
See `crates/stemedb-api/src/extractors.rs` for detailed documentation.
### 3. Register the Route
Add to the `OpenApiRouter` in `crates/stemedb-api/src/lib.rs`:
```rust
let app = OpenApiRouter::new()
.route("/my-endpoint", post(my_handler))
// ...existing routes...
```
The `OpenApiRouter` automatically collects the `#[utoipa::path]` metadata. No separate registration step.
### 4. Verify
```bash
# Build and test
cargo test -p stemedb-api
# Check docs generated correctly
cargo run -p stemedb-api &
curl localhost:18180/api-doc/openapi.json | jq .paths
```
## Error Responses
All error responses use a consistent `ApiError` type that implements `ToSchema`:
```rust
#[derive(Serialize, ToSchema)]
pub struct ApiError {
pub error: String,
pub code: String,
}
```
Map internal errors to API errors in the handler layer. Never leak internal error details.
## Checklist
- [ ] DTO types in `dto.rs` with `Serialize, Deserialize, ToSchema`
- [ ] `From<>` conversions between DTOs and internal types
- [ ] Handler annotated with `#[utoipa::path]`
- [ ] **Query parameters:** Use `QsQuery` if DTO has `Vec` fields, otherwise use `Query`
- [ ] Error responses documented in `responses(...)`
- [ ] Route registered on `OpenApiRouter`
- [ ] Tests cover happy path and error cases
- [ ] `cargo clippy` passes
- [ ] Schema examples are meaningful (not "string")
## Troubleshooting
### OpenAPI spec missing my endpoint
The handler must have `#[utoipa::path]` AND be registered on `OpenApiRouter` (not a plain `axum::Router`).
### Type not showing in schemas
The DTO must derive `ToSchema`. If it references other types, those must also derive `ToSchema`.
### Swagger UI not updating
Hard refresh the browser. The spec is regenerated on each startup.
## Related
- [ai-lookup/patterns/api-documentation.md](../../../ai-lookup/patterns/api-documentation.md) - Toolchain overview
- [ai-lookup/services/api.md](../../../ai-lookup/services/api.md) - Service facts
- [Rust Guidelines](./rust-guidelines.md) - General Rust patterns