Error Model

This chapter is the model underneath Suprnova's error handling - the types, the conversion contract, and the safety guarantees the framework gives you for free. For day-to-day handler patterns (?, returning errors, building custom domain errors) see Error Handling; this chapter explains why those patterns work the way they do.

If you remember one thing from this page: errors in Suprnova are values, not exceptions. Every error eventually becomes an HttpResponse via a single, total conversion. There is no global exception handler because there is no global exception.

The shape

Suprnova's error model has five moving parts:

FrameworkError and the framework's concrete error types use From implementations. A hand-written HttpError must be mapped with FrameworkError::from_http_error before ?; there is no blanket From<T: HttpError> implementation. The middleware chain converts errors at the request boundary, and the panic handler converts an unwind. Ordinary errors then share the common body renderer and 5xx sanitisation rule.

Response is Result<HttpResponse, HttpResponse>

Every handler returns this:

pub type Response = Result<HttpResponse, HttpResponse>;

Both arms carry the same payload type, which is the whole point. When the middleware chain finishes executing your handler it collapses the result with one line:

result.unwrap_or_else(|e| e)

The framework does not need to know whether your handler "succeeded" or "failed" - both arms are already rendered HTTP responses. The distinction exists only so ? can do its job:

use suprnova::{Request, Response, json_response};

pub async fn show(req: Request) -> Response {
    // `?` short-circuits on Err. Each conversion below produces an
    // HttpResponse via a From impl - the chain collapses both arms.
    let id: i64 = req.param("id")?.parse().map_err(|_| {
        suprnova::FrameworkError::param_parse("id", "i64")
    })?;
    let user = User::find_or_fail(id).await?;  // 404 if missing
    Ok(json_response!({ "user": user }))
}

That single contract - every error path produces an HttpResponse through From - is the core of the model. Everything else in this chapter is what the various From impls actually do.

Why Suprnova diverges

Laravel throws exceptions and routes them through a global Handler class registered in app/Exceptions/Handler.php. The framework catches everything, asks the handler "what do I render?", and emits the response. PHP's unwinding-exception model makes this natural.

Rust has no unwinding exceptions in user code. Suprnova's equivalent is the From<FrameworkError> for HttpResponse impl plus the ErrorOccurred event. The conversion is the renderer; the event is where you hook observability (Sentry, PagerDuty, structured shippers). You don't register a handler class - the conversion is a function and listening for ErrorOccurred is the extension point. Same surface, different machinery.

FrameworkError - the canonical enum

Every error path inside the framework - extractors, route binding, the container, validation, the database layer, storage - produces a FrameworkError. It's an enum with seventeen variants, each tagged with its HTTP status:

pub enum FrameworkError {
    ServiceNotFound { type_name: &'static str },        // 500
    ParamError { param_name: String },                   // 400
    ValidationError { field: String, message: String },  // 422
    Database(String),                                    // 500
    Internal { message: String },                        // 500
    Domain { message: String, status_code: u16 },        // *
    Validation(ValidationErrors),                        // 422
    Unauthorized,                                        // 403
    Denial { message: String, status_code: u16, code: String }, // the denial's status, 403 by default
    ModelNotFound { model_name: String },                // 404
    ParamParse { param: String, expected_type: &'static str }, // 400
    UnsupportedMediaType,                                // 415
    PrecognitionSuccess,                                 // 204
    PrecognitionFailure(ValidationErrors),               // 422
    AlreadyReported,                                     // CLI-only
    RateLimited { retry_after: Option<Duration>, message: String }, // 429
    Timeout { elapsed: Duration, message: String },      // 504
    External { message: String, source: Arc<dyn Error + Send + Sync> }, // 500
}

You rarely match on the variant. You construct one through a convenience constructor and let ? do the rest:

use std::time::Duration;
use suprnova::FrameworkError;

// All of these produce a FrameworkError with the right status:
FrameworkError::not_found("User");                    // → ModelNotFound, 404
FrameworkError::bad_request("Bad input");             // → Domain, 400
FrameworkError::param("user_id");                     // → ParamError, 400
FrameworkError::param_parse("user_id", "i64");        // → ParamParse, 400
FrameworkError::validation("email", "required");      // → ValidationError, 422
FrameworkError::domain("Conflict", 409);              // → Domain, 409
FrameworkError::internal("disk full");                // → Internal, 500
FrameworkError::database("timeout");                  // → Database, 500
FrameworkError::timeout(Duration::from_secs(30), "waiting for the export"); // → Timeout, 504

There are no unauthorized() or forbidden() constructors on FrameworkError - Unauthorized is a fixed variant carrying the Laravel "This action is unauthorized." message at 403, and 401 cases go through AppError::unauthorized (next section). Note: the variant is named Unauthorized but the status is 403 because it models Laravel's authorization rejection, not HTTP authentication.

Automatic conversion

FrameworkError implements From<sea_orm::DbErr> and From<opendal::Error> so database and storage errors flow through ? without a wrap:

use suprnova::{DB, FrameworkError};
use sea_orm::ActiveModelTrait;

pub async fn create_user(new_user: ActiveModel) -> Result<Model, FrameworkError> {
    // Both `?` calls here convert into FrameworkError automatically:
    // - DB::get returns Result<_, FrameworkError>
    // - insert returns Result<_, DbErr>, which has From<DbErr> for FrameworkError
    let user = new_user.insert(&*DB::get()?).await?;
    Ok(user)
}

If your code returns Result<_, FrameworkError>, every common error your dependencies produce already speaks the right language. The controller's ? does no work beyond converting one error type into another.

Wrapping context

When you need to re-raise an error with operation context, use .context():

db.insert(user).await
    .map_err(FrameworkError::from)
    .map_err(|e| e.context("creating new user"))?;

The message becomes "creating new user: <original>". The variant is preserved where it matters - Validation, ValidationError, PrecognitionFailure, PrecognitionSuccess, Unauthorized, Denial, ModelNotFound, ParamParse, UnsupportedMediaType, AlreadyReported, RateLimited, Timeout, and External keep their structure so the response renderer still emits the correct shape (and, for External, so the wrapped source survives). Plain message-carrying variants (Internal, Database, Domain) flatten into a Domain with the prefixed message.

Wrapping a foreign error

Every other variant stringifies what it wraps. from_external_with keeps the original error reachable, so logs can render the whole chain and code can still ask what actually failed:

use suprnova::FrameworkError;

let row = sqlx_like_query()
    .await
    .map_err(|e| FrameworkError::from_external_with("verify query failed", e))?;

from_external(e) is the same thing with the error's own Display as the message. Both map to HTTP 500.

To inspect the original, use external_source() rather than source():

if let Some(src) = err.external_source() {
    if let Some(db) = src.downcast_ref::<sea_orm::DbErr>() {
        // decide whether this is worth retrying
    }
}

std::error::Error::source() hands back the shared Arc handle, not the wrapped error, so downcasting through it returns None. external_source() dereferences the handle first.

The framework renders the full chain into the 5xx log line and into the debug_message field it adds when APP_DEBUG=true, so a wrapped error's text is never lost.

Preserving rate-limit hints

RateLimited exists so a downstream Retry-After hint survives the trip through the error system as a Duration instead of collapsing into message text:

use std::time::Duration;
use suprnova::FrameworkError;

let err = FrameworkError::rate_limited(
    Some(Duration::from_secs(30)),
    "push provider rejected the batch",
);

assert_eq!(err.retry_after(), Some(Duration::from_secs(30)));
assert_eq!(err.status_code(), 429);

retry_after() returns None for every other variant and for throttles that arrived without a hint. The variant renders as HTTP 429, and .context(...) preserves it rather than flattening to Domain, so the duration is never stripped by adding operation context.

Telling a passed deadline from a failure

Timeout says that a deadline passed before the awaited work finished. It carries the deadline as elapsed and what was awaited as message. The workflow wait returns it: WorkflowHandle::wait_with_timeout and wait_with_options end with FrameworkError::Timeout when the workflow is still pending or running at the deadline (see Workflows). A failed status query is a different error, so a caller can match the two apart:

use std::time::Duration;
use suprnova::FrameworkError;

let err = FrameworkError::timeout(Duration::from_secs(30), "waiting for the export");

assert!(err.is_timeout());
assert_eq!(err.status_code(), 504);
assert_eq!(err.to_string(), "Timed out after 30s: waiting for the export");

The error does not cancel the work. It renders as HTTP 504, and like every 5xx the client sees the generic Internal Server Error message while the detail goes to the logs. .context(...) preserves the variant, so a caller further up can still match on it.

AppError - ad-hoc domain errors

For one-off errors where you don't want to define a dedicated type, use AppError. It implements HttpError and has a From into FrameworkError, so ? works directly:

use suprnova::{AppError, Request, Response, json_response};

pub async fn transfer(req: Request) -> Response {
    let amount: i64 = req.param("amount")?.parse()
        .map_err(|_| AppError::bad_request("amount must be a number"))?;

    if amount <= 0 {
        return Err(AppError::unprocessable("amount must be positive").into());
    }

    if amount > 1_000_000 {
        return Err(AppError::forbidden("amount exceeds daily limit").into());
    }

    Ok(json_response!({ "transferred": amount }))
}

The constructors map cleanly onto Laravel's abort($status, $msg) shape:

Note AppError::unauthorized is 401 (HTTP authentication missing), while FrameworkError::Unauthorized is 403 (authorization denied, matching Laravel's policy rejection). They mean different things; pick the one that matches the failure.

HttpError - custom typed errors

When the same domain error appears in many places, model it as a type. Implement HttpError and the conversion is yours:

use suprnova::HttpError;

#[derive(Debug)]
pub struct InsufficientFunds {
    pub available: i64,
    pub requested: i64,
}

impl std::fmt::Display for InsufficientFunds {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        write!(f, "Insufficient funds: have {}, need {}",
            self.available, self.requested)
    }
}

impl std::error::Error for InsufficientFunds {}

impl HttpError for InsufficientFunds {
    fn status_code(&self) -> u16 { 402 }
    fn error_message(&self) -> String {
        format!("Need {} units, only {} available.",
            self.requested, self.available)
    }
}

HttpError has two methods, both with defaults:

pub trait HttpError: std::error::Error + Send + Sync + 'static {
    fn status_code(&self) -> u16 { 500 }
    fn error_message(&self) -> String { self.to_string() }
}

Bridging to ?

A naive impl<T: HttpError> From<T> for FrameworkError would conflict with the existing From<AppError> impl (because AppError itself implements HttpError). Suprnova resolves the orphan-rule problem with an explicit bridge constructor instead:

use suprnova::{FrameworkError, HttpError};

pub async fn debit(account: &mut Account, amount: i64) -> Result<(), FrameworkError> {
    account.withdraw(amount)
        .map_err(FrameworkError::from_http_error)?;
    Ok(())
}

The status code and message are taken from HttpError::status_code and HttpError::error_message and stored in a FrameworkError::Domain variant. The response renderer then follows the normal Domain path.

#[domain_error] for boilerplate-free types

If you want the typed-error pattern without writing the Display, Error, and HttpError impls by hand, use the #[domain_error] attribute macro:

use suprnova::domain_error;

#[domain_error(status = 404, message = "User not found")]
pub struct UserNotFoundError;

#[domain_error(status = 402, message = "Insufficient funds")]
pub struct InsufficientFundsError {
    pub available: i64,
    pub requested: i64,
}

#[domain_error] generates the full impl set including From<YourError> for FrameworkError, so ? works directly with no bridge call:

pub async fn show(req: Request) -> Response {
    let id: i64 = req.param("id")?.parse()
        .map_err(|_| FrameworkError::param_parse("id", "i64"))?;
    let user = User::find(id).await?
        .ok_or_else(|| FrameworkError::from(UserNotFoundError))?;
    Ok(json_response!({ "user": user }))
}

The three tiers of custom error story - AppError for inline, #[domain_error] for typed-with-macro, hand-rolled HttpError for full control - give you the right tool at every level of formality.

ValidationErrors - the Laravel-shaped error bag

When a request fails validation, Suprnova emits the same JSON shape Laravel and Inertia front-ends expect:

{
    "message": "The given data was invalid.",
    "errors": {
        "email": ["The email field must be a valid email address."],
        "password": ["The password must be at least 8 characters."]
    },
    "request_id": "8f9e1a2b-c3d4-..."
}

You usually don't build this by hand - #[derive(Validate)] on a form request and the validator crate behind it produces a validator::ValidationErrors which Suprnova converts via ValidationErrors::from_validator. But the type is public when you need it:

use suprnova::{FrameworkError, ValidationErrors};

pub async fn after_validation(payload: &Signup) -> Result<(), FrameworkError> {
    let mut errs = ValidationErrors::new();

    if payload.email.ends_with("@example.com") {
        errs.add("email", "example.com addresses are not allowed");
    }
    if payload.password == payload.email {
        errs.add("password", "password must not match email");
    }

    errs.into_result().map_err(FrameworkError::Validation)
}

add_to_bag scopes errors under a named bag (Laravel's withErrors($errors, 'profile') shape) by prepending the bag name with a . separator:

let mut errs = ValidationErrors::new();
errs.add_to_bag("profile", "bio", "must be under 280 characters");
errs.add_to_bag("billing", "card", "expired");
// errors map: { "profile.bio": [...], "billing.card": [...] }

retain_fields keeps only the listed entries - used internally by Precognition's Precognition-Validate-Only header so the server runs full validation but reports errors only for the fields the client asked about.

The conversion contract

When a FrameworkError reaches an HTTP boundary it goes through From<FrameworkError> for HttpResponse. Three things happen, in order:

1. Status routing. The variant's status_code() is read once.
2. Logging + observability. 5xx fires tracing::error! and dispatches ErrorOccurred; 4xx fires tracing::warn!. Both carry the request id when one is in scope.
3. Body rendering. A JSON body in the Laravel shape, sanitised for 5xx.

The ordinary body shape

Ordinary error responses that reach the common renderer follow this JSON skeleton:

{
    "message": "<human readable>",
    "errors": { "field": ["msg", ...] },
    "request_id": "<uuid>" | null,
    "debug_message": "<dev only>"
}

• message is always present on these ordinary responses.
• errors only appears for validation-style errors (Validation, ValidationError) - both render the same shape so consumers parse one path.
• request_id appears on these ordinary responses (null when outside a request scope, such as during early boot or in tests with no request context).
• debug_message only appears for 5xx when APP_DEBUG=true. It is strictly additive - production clients must not key on it. With debug on, a browser or an Inertia visit gets the development error page in place of this body.

Three special variants return before request-id injection:

• PrecognitionSuccess is a bodyless 204 response.
• PrecognitionFailure contains the validation body plus Precognition headers.
• An accidentally HTTP-rendered AlreadyReported sentinel is a generic 500 response containing only message.

The 5xx sanitisation rule

This is the safety guarantee worth memorising. For any error with status ≥ 500 that reaches the common renderer, the JSON body's message is replaced with the literal string:

{ "message": "Internal Server Error", "request_id": "..." }

The raw error detail does not leak to the response body. It goes to:

• the tracing::error! log entry, with the request id and status
• the ErrorOccurred event, which any listener can pick up

When APP_DEBUG=true (false by default outside local/dev/test), the response also carries a debug_message field with the raw detail - but message stays generic in both modes, so frontends and clients can't accidentally couple to dev-only data. A browser or an Inertia visit gets the development error page instead, which shows the same detail as HTML.

This is the contract that lets you call FrameworkError::internal("db connection refused: password mismatch on user 'app_rw'") without leaking the password to the wire. The message you pass is for operators reading logs; the message the client sees is "Internal Server Error".

For 4xx errors, the caller-facing message is preserved - 404 User not found, 400 Missing required parameter: user_id. These are domain errors the client needs to act on, not internal failures.

Where the contract lives

The whole conversion is one function - impl From<FrameworkError> for HttpResponse in framework/src/http/response.rs. Read it once and you've read the entire error rendering surface of Suprnova. There is no other path.

The panic boundary

A panic in a middleware or handler would otherwise propagate up the per-connection task and tear down the hyper service mid-response, leaving the client with a TCP reset and no HTTP response. Suprnova catches it.

execute_chain_safely in framework/src/server.rs wraps the middleware chain in AssertUnwindSafe(...).catch_unwind().await. On a panic it:

1. Extracts the panic payload (handles &'static str and String payloads; anything else surfaces as "panic with non-string payload").
2. Logs tracing::error! with the request method, path, and id.
3. Constructs FrameworkError::internal(format!("request handler panicked: {msg}")) and routes it through the same From<FrameworkError> for HttpResponse conversion every other 5xx uses.
4. Echoes the request id back as X-Request-Id.

The panic payload stays in the log entry; the client gets the sanitised {"message": "Internal Server Error"} body. With debug on, a browser or an Inertia visit gets the development error page instead, with the panic message and the line that panicked. Observability listeners that fire on ErrorOccurred for returned 5xx errors also fire on panics - there is no separate panic-event surface to wire up.

The same panic-recovery pattern is used by:

• WebSocket handlers (framework/src/server.rs)
• Scheduled tasks (framework/src/schedule/mod.rs)
• Workflows (framework/src/workflow/mod.rs)
• The Supervisor trait (broadcasting)

A panic in one of these subsystems is logged and either translated to an error state or auto-restarted; it does not bring down the worker task.

The development error page

With debug mode on, a failed request is hard to read as a debug_message string in a browser tab. So for a browser or an Inertia visit, Suprnova replaces a 5xx built from a failure with the development error page: one HTML document that shows what failed, where, and on which request.

When it renders

The server replaces a response with the page when all of these hold:

• Config::is_debug() returns true: APP_DEBUG=true, or APP_DEBUG unset in the local, development, or testing environment.
• The status is 500 or above.
• The response carries an ErrorReport: a FrameworkError became the response, the panic boundary caught a panic, or one of the framework's middleware answered a failure with a 5xx.
• The request is an Inertia visit (X-Inertia: true), or its Accept header lists text/html.

Everything else stays as it is. A 4xx keeps its body, and so does a 5xx your handler built itself without an error. A client whose Accept header doesn't list text/html (application/json, */*, or no Accept at all) and isn't an Inertia visit keeps the JSON body with debug_message. With debug off, you receive a minimal HTML view for a 5xx when your request accepts text/html and is not an Inertia visit. You see its status without error details or frames. Your Inertia visits keep the application error page, and your JSON requests keep their JSON body. No frames or request details are recorded for this production view.

The page is built in the server, after the panic boundary, so it covers panics as well as returned errors. For these responses it takes the place of the app's Inertia error page. It keeps the status, the error report, Set-Cookie, X-Request-Id, and the other headers that describe the request, and it replaces the ones that described the old body.

To see it, run the app with APP_DEBUG=true and open a route that fails in a browser. From a terminal, ask for HTML:

curl -H 'Accept: text/html' http://localhost:8000/invoices/42

What it shows

• The type of what failed, above the headline, as Laravel's page shows the exception class. An error built with FrameworkError::from_external or from_external_with is named by the type it wraps, as std::any::type_name spells it: std::io::error::Error for an std::io::Error. Any other FrameworkError is named by its variant, such as FrameworkError::ModelNotFound, and a panic is named panic. ErrorReport::type_name() returns the same name.
• The error chain, one line per source, or the panic message and the file:line:column where the panic was raised.
• The site that created the error: the line of the call to a FrameworkError or AppError constructor, or of a ? that converts the error.
• The stack frames recorded at that site. Your application's frames are listed: your handlers and middleware, and the code they call. Each run of frames from the framework, the async runtime, other dependencies, and the standard library is collapsed behind a count. You see the source line and a few nearby lines for every application frame whose source file is readable. Each source error with recorded frames has its own trace.
• The request: the method, the path, the query parameters, the headers, the matched route pattern, and the request id.

Frames are recorded for an error created on the request's own task by a FrameworkError or AppError constructor or by a From conversion, which includes ?. An error built as a struct literal, or created on a task the request spawned, has no frames, and the page says so.

The constructors and conversions are #[track_caller], so the site is the line that called them, even in a build without debug info. A call made inside a function that is not #[track_caller] reports that function's line instead: with .map_err(FrameworkError::from)?, the site is a line of Result::map_err in the Rust toolchain, and the frames show the handler that called it. A build without debug info names each frame's function but not its line. The frame that created the error is the exception: it shows the creation site as its location.

Open a frame in your editor

Set APP_EDITOR and every frame with a location links to its line in your editor, as Laravel's app.editor does:

APP_EDITOR=vscode

• A name Laravel knows uses that editor's URL format. vscode links to vscode://file/{file}:{line} and phpstorm to phpstorm://open?file={file}&line={line}. The other names are antigravity, atom, cursor, emacs, fleet, idea, kiro, macvim, neovim, netbeans, nova, sublime, textmate, trae, vscode-insiders, vscode-insiders-remote, vscode-remote, vscodium, windsurf, xdebug, and zed.
• A value that holds {file} is a template: APP_EDITOR=myeditor://{file}#{line}.
• Any other value name links to name://open?file={file}&line={line}.

{file} is the absolute path of the frame's file. A relative location is resolved the way the page finds source lines: from the working directory first. {line} is the frame's line. Without APP_EDITOR, no frame carries a link.

What it never shows

• The values of the Authorization, Proxy-Authorization, Cookie, and Set-Cookie headers, and of every header and query parameter whose name contains token, secret, password, key, or signature in any letter case. Each value reads [redacted].
• The value of every such parameter anywhere else in the text the page shows: in a Referer or an X-Original-URI header that repeats the request's URL, /reset?token=... reads /reset?token=[redacted].
• The password of a URL with credentials: postgres://app:hunter2@db/app reads postgres://app:[redacted]@db/app.
• The request body, environment variables, and configuration values. The page never reads them.

What it loads

Nothing. The page is one HTML document with an inline style sheet and no JavaScript. It requests no script, style sheet, font, image, or frame, so it renders when the frontend build, the Vite manifest, Inertia, or the view layer is what failed. Every value it shows is HTML-escaped. The response carries Cache-Control: no-store and a Content-Security-Policy of default-src 'none'; style-src 'unsafe-inline'; base-uri 'none'; form-action 'none', which allows no script.

Why Suprnova diverges

Laravel's exception page shows the request body, lists the SQL queries the request ran, and runs a bundled JavaScript app that adds a "copy as Markdown" button. Suprnova's page differs on all three:

• It never shows the request body. A body often holds the passwords, tokens, and personal data the header redaction exists to protect, and a development server on a shared network or behind a tunnel shows the page to whoever triggers the error.
• It runs no script and loads nothing, so it renders in the situations where a developer needs it most: a broken frontend build, a missing Vite manifest, or a failure inside Inertia.
• It lists no queries. SQL debugging is separate work, outside this page.

Hooking observability with ErrorOccurred

ErrorOccurred is a built-in event the framework dispatches on every 5xx response (including the ones synthesised from panics):

pub struct ErrorOccurred {
    pub error_message: String,
    pub status_code: u16,
    pub request_id: Option<String>,
}

Listen for it the same way you listen for any event:

use std::sync::Arc;
use suprnova::{ErrorOccurred, EventFacade, FrameworkError, Listener};

pub struct SentryReporter;

#[suprnova::async_trait]
impl Listener<ErrorOccurred> for SentryReporter {
    async fn handle(&self, evt: &ErrorOccurred) -> Result<(), FrameworkError> {
        sentry::capture_message(&evt.error_message, sentry::Level::Error);
        Ok(())
    }
}

// In bootstrap.rs:
EventFacade::listen::<ErrorOccurred, _>(Arc::new(SentryReporter)).await;

This is the Suprnova equivalent of Laravel's report() callback on the global exception handler. The event arrives with the original unsanitised error_message (the body the client sees is still sanitised), the status code, and the correlatable request id.

Rendering the full chain: render_error_chain

thiserror's generated Display prints only an error's own message, so a FrameworkError::External's wrapped source is invisible unless something walks the chain. render_error_chain does that walk and joins the result with ": ", the same separator .context() uses - the framework calls it before building error_message above and before the matching 5xx log line, which is why a wrapped error doesn't lose its cause in either place.

Reach for it yourself when a listener or a log sink needs the same full-chain rendering, for example re-wrapping error_message before forwarding it to a sink that only takes a flat string:

use suprnova::render_error_chain;

let chain = render_error_chain(&err);
// "loading users: connection refused (os error 111)"

Abort helpers

Three free functions short-circuit a handler at a given status. They mirror Laravel's abort / abort_if / abort_unless:

use suprnova::{abort_with, abort_if, abort_unless, Auth, Request, Response, json_response};

pub async fn show(req: Request) -> Response {
    abort_unless(Auth::check(), 401, "must be logged in")?;
    abort_if(req.param("id")? == "0", 404, "User not found")?;
    abort_with(503, "scheduled maintenance")?;
    Ok(json_response!({ "ok": true }))
}

Each returns Result<(), FrameworkError>. Use them with ?. The underlying error is FrameworkError::Domain { message, status_code }, so it renders through the same body shape and sanitisation rules as every other error. Out-of-range status codes are coerced to 500 by the response renderer's status validation; you don't need to defend against bad input at the call site.

The CLI sentinel: AlreadyReported

One variant of FrameworkError has no HTTP meaning. AlreadyReported is constructed via FrameworkError::silent() and used by the console dispatcher when clap has already formatted and printed its own argument-parse error. The binary's main translates the sentinel to a non-zero exit code without eprintln, so users never see two error messages for the same failure.

If AlreadyReported ever reaches an HTTP response converter, it indicates a request handler accidentally returned silent(). The converter logs a loud tracing::error! identifying the leak and returns a generic 500 containing only {"message": "Internal Server Error"}. The variant has no business in the request path, and the loud log makes the bug observable instead of silent.

You don't normally see this variant; it's documented here because the enum is HTTP-flavoured and the otherwise-unexplained variant would puzzle anyone reading the source.

Safety guarantees, in summary

The contract Suprnova gives you:

• Total conversion. Every FrameworkError produces an HttpResponse. There is no error path that crashes the server or drops the connection silently.
• Sanitised 5xx. The common renderer replaces the wire message for any 5xx with Internal Server Error; raw detail flows to logs and ErrorOccurred. An accidentally HTTP-rendered AlreadyReported sentinel returns the same generic message without request_id.
• Optional debug visibility. APP_DEBUG=true adds a debug_message field for ordinary 5xx responses, never message. Production clients cannot accidentally couple to dev-only data. A browser or an Inertia visit gets the development error page instead, which redacts credentials and never shows the request body, environment variables, or configuration values.
• Correlatable request ids. Every ordinary error body that reaches the common renderer carries the request id (or null when no request scope exists); the same id appears in the log line and the ErrorOccurred event. The three early-return variants described above bypass this field.
• Panic recovery. Panics in handlers and middleware are caught, logged, and routed through the same From impl as returned errors. No connection drop, no observability gap.
• One common shape for ordinary errors. Validation errors, parameter errors, panics, custom domain errors, and storage failures that reach the common renderer use the same JSON skeleton. The three special variants documented above have distinct wire shapes.

Where each piece lives

Next

• Error Handling - the practical handler patterns that use this model
• Request Lifecycle - where in the request flow the error conversion runs
• Validation - #[derive(Validate)], form requests, and how ValidationErrors gets populated
• Responses - HttpResponse builders, headers, cookies, streaming
• Events - listening to ErrorOccurred and other built-in events