harn_vm/value/
error.rs

1use harn_lexer::Span;
2
3use super::{VmDictExt, VmValue};
4
5/// Bound expressing how many arguments a callable accepts. Used in
6/// [`VmError::ArityMismatch`] so error messages can render the exact
7/// signature contract the caller violated.
8#[derive(Debug, Clone, Copy, PartialEq, Eq)]
9pub enum ArityExpect {
10    /// Exactly N parameters, no defaults, no rest.
11    Exact(usize),
12    /// `min..=max`: some params have defaults but the upper bound is fixed.
13    Range { min: usize, max: usize },
14    /// At least N parameters; further args land in a rest list. Used for
15    /// `print` / `log` / variadics.
16    AtLeast(usize),
17}
18
19impl std::fmt::Display for ArityExpect {
20    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
21        match self {
22            ArityExpect::Exact(n) => write!(f, "{n}"),
23            ArityExpect::Range { min, max } => write!(f, "{min}..={max}"),
24            ArityExpect::AtLeast(n) => write!(f, "at least {n}"),
25        }
26    }
27}
28
29#[derive(Debug, Clone)]
30pub struct ArityMismatchError {
31    pub callee: String,
32    pub expected: ArityExpect,
33    pub got: usize,
34    pub span: Option<Span>,
35}
36
37#[derive(Debug, Clone, Copy, PartialEq, Eq)]
38pub enum DeadlockDiagnostic {
39    SelfDeadlock,
40    WaitForGraph,
41}
42
43impl DeadlockDiagnostic {
44    fn code(self) -> &'static str {
45        match self {
46            Self::SelfDeadlock => "HARN-ORC-011",
47            Self::WaitForGraph => "HARN-ORC-012",
48        }
49    }
50}
51
52/// Payload for [`VmError::Deadlock`]. `kind` is the primitive kind
53/// (`"mutex"`, `"channel"`) or `"task"`; `key` is the primitive key or task
54/// id; `detail` names the specific footgun.
55#[derive(Debug, Clone)]
56pub struct DeadlockError {
57    pub diagnostic: DeadlockDiagnostic,
58    pub kind: String,
59    pub key: String,
60    pub detail: String,
61}
62
63impl DeadlockError {
64    pub(crate) fn self_deadlock(
65        kind: impl Into<String>,
66        key: impl Into<String>,
67        detail: impl Into<String>,
68    ) -> Self {
69        Self {
70            diagnostic: DeadlockDiagnostic::SelfDeadlock,
71            kind: kind.into(),
72            key: key.into(),
73            detail: detail.into(),
74        }
75    }
76
77    pub(crate) fn wait_for_graph(
78        kind: impl Into<String>,
79        key: impl Into<String>,
80        detail: impl Into<String>,
81    ) -> Self {
82        Self {
83            diagnostic: DeadlockDiagnostic::WaitForGraph,
84            kind: kind.into(),
85            key: key.into(),
86            detail: detail.into(),
87        }
88    }
89}
90
91#[derive(Debug, Clone)]
92pub struct ArgTypeMismatchError {
93    pub callee: String,
94    pub param: String,
95    pub expected: String,
96    pub got: &'static str,
97    pub span: Option<Span>,
98}
99
100#[derive(Debug, Clone)]
101pub enum VmError {
102    StackUnderflow,
103    StackOverflow,
104    UndefinedVariable(String),
105    UndefinedBuiltin(String),
106    ImmutableAssignment(String),
107    TypeError(String),
108    Runtime(String),
109    DivisionByZero,
110    /// A host-imposed deadline expired while executing the VM. Unlike a Harn
111    /// `deadline` block, this control-plane stop cannot be caught by user code.
112    ExecutionDeadlineExceeded,
113    /// A Harn program requested that its embedding process terminate with this
114    /// status code. Like a host deadline, this is control flow rather than a
115    /// catchable Harn error; the embedding boundary owns the final cleanup and
116    /// process exit.
117    ProcessExit(i32),
118    /// A host dropped a polled top-level execution future. Interpreter state is
119    /// intentionally not resumed after arbitrary async cancellation. Discard
120    /// this VM; see [`crate::Vm::execute_with_timeout`] for ambient-state
121    /// cleanup requirements.
122    AbandonedExecution,
123    Thrown(VmValue),
124    /// Thrown with error category for structured error handling.
125    CategorizedError {
126        message: String,
127        category: ErrorCategory,
128    },
129    DaemonQueueFull {
130        daemon_id: String,
131        capacity: usize,
132    },
133    /// A deterministic, provably-unresolvable self-deadlock caught before the
134    /// VM would block forever (Rust's borrow checker prevents data races but
135    /// not deadlocks; this is the Go-runtime "all goroutines asleep" analogue
136    /// for the cases we can prove). Boxed — like [`VmError::ArityMismatch`] —
137    /// so the rare three-`String` payload doesn't enlarge `VmError` on the
138    /// pervasive `Result<VmValue, VmError>` hot path. Carries `HARN-ORC-011`.
139    Deadlock(Box<DeadlockError>),
140    Return(VmValue),
141    InvalidInstruction(u8),
142    /// Wrong number of arguments at a call site. Distinct from
143    /// [`VmError::TypeError`] so the runtime can match-and-recover (and
144    /// so error UX renders `expected 2..=3 got 1` consistently).
145    ArityMismatch(Box<ArityMismatchError>),
146    /// Argument value did not satisfy the declared parameter type.
147    /// `expected` is a pretty-printed type expression; `got` is the value's
148    /// runtime type name (`VmValue::type_name`). Used for both
149    /// user-defined function parameters (with declared types) and
150    /// registry-known builtin parameters.
151    ArgTypeMismatch(Box<ArgTypeMismatchError>),
152}
153
154impl VmError {
155    /// Whether this error is VM control flow that user `catch` blocks and
156    /// error-as-data combinators must propagate unchanged.
157    pub fn is_uncatchable_control_flow(&self) -> bool {
158        matches!(self, Self::ExecutionDeadlineExceeded | Self::ProcessExit(_))
159    }
160
161    /// The requested host-process exit status, when this is an explicit Harn
162    /// `exit(code)` control signal.
163    pub fn process_exit_code(&self) -> Option<i32> {
164        match self {
165            Self::ProcessExit(code) => Some(*code),
166            _ => None,
167        }
168    }
169
170    /// The `VmValue` a `catch` binding (or a `parallel settle` result) observes
171    /// for this error: the raw thrown value for [`VmError::Thrown`] (so a
172    /// structured error — e.g. a `{category, message}` dict from `throw_error` —
173    /// keeps its shape and category), a structured `{category, message}` dict for
174    /// [`VmError::CategorizedError`] (so consumers branch on the typed category
175    /// rather than substring-matching rendered prose), otherwise the rendered
176    /// message.
177    ///
178    /// Single source of truth for VM-error-to-value lowering so every seam that
179    /// surfaces a caught error to Harn (`try`/`catch` via `handle_error`,
180    /// `parallel settle`) exposes identical, structure-preserving values. Before
181    /// this was shared, `parallel settle` stringified errors via `to_string()`,
182    /// so a categorized error thrown in a settle branch lost its category (a
183    /// `cancelled`/`internal` fault that must propagate looked `generic`).
184    ///
185    /// The `category` key uses [`ErrorCategory::as_str`] — a canonical,
186    /// exhaustively-matched snake_case contract — so a new variant added there
187    /// is compiler-forced to name its key. `message` preserves the original
188    /// rendered text, so a `catch` that stringifies the caught value still reads
189    /// sensibly (the dict renders both fields).
190    pub fn thrown_value(&self) -> VmValue {
191        match self {
192            VmError::Thrown(v) => v.clone(),
193            VmError::CategorizedError { message, category } => {
194                let mut dict = std::collections::BTreeMap::new();
195                dict.put_str("category", category.as_str());
196                dict.put_str("message", message);
197                VmValue::dict(dict)
198            }
199            other => VmValue::String(arcstr::ArcStr::from(other.to_string())),
200        }
201    }
202}
203
204/// Error categories for structured error handling in agent orchestration.
205#[derive(Debug, Clone, PartialEq, Eq)]
206pub enum ErrorCategory {
207    /// Network/connection timeout
208    Timeout,
209    /// Authentication/authorization failure
210    Auth,
211    /// Rate limit exceeded (HTTP 429 / quota)
212    RateLimit,
213    /// Upstream provider is overloaded (HTTP 503 / 529).
214    /// Distinct from RateLimit: the client hasn't exceeded a quota — the
215    /// provider is shedding load and will recover on its own.
216    Overloaded,
217    /// Provider-side 5xx error (500, 502) that isn't specifically overload.
218    ServerError,
219    /// Network-level transient failure (connection reset, DNS hiccup,
220    /// partial stream) — retryable but not provider-status-coded.
221    TransientNetwork,
222    /// LLM output failed schema validation. Retryable via `schema_retries`.
223    SchemaValidation,
224    /// LLM streaming response was aborted mid-stream because the partial
225    /// JSON content could not conceivably satisfy `output_schema`. Surfaced
226    /// by `llm_call` when `schema_stream_abort` is on (the default for
227    /// schema-bearing calls). Consumes one `schema_retries` budget slot;
228    /// the retry replays the prompt with a corrective nudge that cites
229    /// the abort path + reason.
230    SchemaStreamAborted,
231    /// Tool execution failure
232    ToolError,
233    /// Tool was rejected by the host (not permitted / not in allowlist)
234    ToolRejected,
235    /// Outbound network egress was blocked by policy.
236    EgressBlocked,
237    /// Operation was cancelled
238    Cancelled,
239    /// Channel was closed before the operation could complete.
240    ChannelClosed,
241    /// Resource not found
242    NotFound,
243    /// Circuit breaker is open
244    CircuitOpen,
245    /// LLM cost or token budget would be exceeded
246    BudgetExceeded,
247    /// An internal engine/wiring bug — an undefined builtin, corrupt bytecode,
248    /// or another VM invariant violation that no amount of retrying or model
249    /// reasoning can fix. Distinct from `Generic` so callers (notably the agent
250    /// loop) can re-raise it loudly instead of folding it into a tool-error
251    /// observation and marching on to a `done` status. This is the category
252    /// that keeps a mis-wired builtin (e.g. a `#[harn_builtin]` def missing
253    /// from its install array) from shipping silently inert.
254    Internal,
255    /// Generic/unclassified error
256    Generic,
257}
258
259impl ErrorCategory {
260    pub fn as_str(&self) -> &'static str {
261        match self {
262            ErrorCategory::Timeout => "timeout",
263            ErrorCategory::Auth => "auth",
264            ErrorCategory::RateLimit => "rate_limit",
265            ErrorCategory::Overloaded => "overloaded",
266            ErrorCategory::ServerError => "server_error",
267            ErrorCategory::TransientNetwork => "transient_network",
268            ErrorCategory::SchemaValidation => "schema_validation",
269            ErrorCategory::SchemaStreamAborted => "schema_stream_aborted",
270            ErrorCategory::ToolError => "tool_error",
271            ErrorCategory::ToolRejected => "tool_rejected",
272            ErrorCategory::EgressBlocked => "egress_blocked",
273            ErrorCategory::Cancelled => "cancelled",
274            ErrorCategory::ChannelClosed => "channel_closed",
275            ErrorCategory::NotFound => "not_found",
276            ErrorCategory::CircuitOpen => "circuit_open",
277            ErrorCategory::BudgetExceeded => "budget_exceeded",
278            ErrorCategory::Internal => "internal",
279            ErrorCategory::Generic => "generic",
280        }
281    }
282
283    pub fn parse(s: &str) -> Self {
284        match s {
285            "timeout" => ErrorCategory::Timeout,
286            "auth" => ErrorCategory::Auth,
287            "rate_limit" => ErrorCategory::RateLimit,
288            "overloaded" => ErrorCategory::Overloaded,
289            "server_error" => ErrorCategory::ServerError,
290            "transient_network" => ErrorCategory::TransientNetwork,
291            "schema_validation" => ErrorCategory::SchemaValidation,
292            "schema_stream_aborted" => ErrorCategory::SchemaStreamAborted,
293            "tool_error" => ErrorCategory::ToolError,
294            "tool_rejected" => ErrorCategory::ToolRejected,
295            "egress_blocked" => ErrorCategory::EgressBlocked,
296            "cancelled" => ErrorCategory::Cancelled,
297            "channel_closed" => ErrorCategory::ChannelClosed,
298            "not_found" => ErrorCategory::NotFound,
299            "circuit_open" => ErrorCategory::CircuitOpen,
300            "budget_exceeded" => ErrorCategory::BudgetExceeded,
301            "internal" => ErrorCategory::Internal,
302            _ => ErrorCategory::Generic,
303        }
304    }
305
306    /// Whether this category represents an internal engine/wiring bug that must
307    /// be surfaced rather than retried or swallowed as a recoverable failure.
308    pub fn is_internal(&self) -> bool {
309        matches!(self, ErrorCategory::Internal)
310    }
311
312    /// Whether an error of this category is worth retrying for a transient
313    /// provider-side reason. Agent loops consult this to decide whether to
314    /// back off and retry vs surface the error to the user.
315    pub fn is_transient(&self) -> bool {
316        matches!(
317            self,
318            ErrorCategory::Timeout
319                | ErrorCategory::RateLimit
320                | ErrorCategory::Overloaded
321                | ErrorCategory::ServerError
322                | ErrorCategory::TransientNetwork
323        )
324    }
325}
326
327/// Create a categorized error conveniently.
328pub fn categorized_error(message: impl Into<String>, category: ErrorCategory) -> VmError {
329    VmError::CategorizedError {
330        message: message.into(),
331        category,
332    }
333}
334
335/// Extract error category from a VmError.
336///
337/// Classification priority:
338/// 1. Explicit CategorizedError variant (set by throw_error or internal code)
339/// 2. Thrown dict with a "category" field (user-created structured errors)
340/// 3. HTTP status code extraction (standard, unambiguous)
341/// 4. Deadline exceeded (VM-internal)
342/// 5. Fallback to Generic
343pub fn error_to_category(err: &VmError) -> ErrorCategory {
344    match err {
345        VmError::ExecutionDeadlineExceeded => ErrorCategory::Timeout,
346        // ProcessExit is uncatchable control flow rather than an agent-facing
347        // failure. Keep this fallback total for callers that classify an
348        // arbitrary VmError without treating the request as retryable.
349        VmError::ProcessExit(_) => ErrorCategory::Generic,
350        VmError::AbandonedExecution => ErrorCategory::Cancelled,
351        VmError::CategorizedError { category, .. } => category.clone(),
352        VmError::Thrown(VmValue::Dict(d)) => d
353            .get("category")
354            .map(|v| ErrorCategory::parse(&v.display()))
355            .unwrap_or(ErrorCategory::Generic),
356        VmError::Thrown(VmValue::String(s)) => classify_error_message(s),
357        VmError::Runtime(msg) => classify_error_message(msg),
358        // Engine/wiring bugs: an undefined builtin (declared but not installed,
359        // or a typo in stdlib/host code) or corrupt bytecode. No retry or model
360        // reasoning fixes these, so they get their own category the agent loop
361        // re-raises instead of swallowing.
362        VmError::UndefinedBuiltin(_) | VmError::InvalidInstruction(_) => ErrorCategory::Internal,
363        // A deadlock is permanently non-retryable and not provider-related —
364        // `Generic` is the correct "surface it, don't back off" bucket.
365        VmError::Deadlock(_) => ErrorCategory::Generic,
366        _ => ErrorCategory::Generic,
367    }
368}
369
370/// Classify an error message using HTTP status codes and well-known patterns.
371/// Prefers unambiguous signals (status codes) over substring heuristics.
372pub fn classify_error_message(msg: &str) -> ErrorCategory {
373    // 1. HTTP status codes — most reliable signal
374    if let Some(cat) = classify_by_http_status(msg) {
375        return cat;
376    }
377    // 2. Internal engine/wiring bug surfaced as a plain message. Some call
378    //    sites build `Runtime("Undefined builtin: …")` strings instead of the
379    //    structured `VmError::UndefinedBuiltin` variant; classify both the same
380    //    so the agent loop re-raises rather than swallows.
381    if msg.contains("Undefined builtin") {
382        return ErrorCategory::Internal;
383    }
384    // 3. Well-known error identifiers from major APIs
385    //    (Anthropic, OpenAI, and standard HTTP patterns)
386    let lower = msg.to_lowercase();
387    if lower.contains("cancelled") || lower.contains("canceled") {
388        return ErrorCategory::Cancelled;
389    }
390    if msg.contains("ChannelClosed") || lower.contains("channel closed") {
391        return ErrorCategory::ChannelClosed;
392    }
393    if msg.contains("Deadline exceeded") || msg.contains("context deadline exceeded") {
394        return ErrorCategory::Timeout;
395    }
396    if msg.contains("overloaded_error") {
397        // Anthropic overloaded_error surfaces as HTTP 529.
398        return ErrorCategory::Overloaded;
399    }
400    if msg.contains("api_error") {
401        // Anthropic catch-all server-side error.
402        return ErrorCategory::ServerError;
403    }
404    if msg.contains("insufficient_quota") || msg.contains("billing_hard_limit_reached") {
405        // OpenAI-specific quota error types.
406        return ErrorCategory::RateLimit;
407    }
408    if msg.contains("invalid_api_key") || msg.contains("authentication_error") {
409        return ErrorCategory::Auth;
410    }
411    if msg.contains("not_found_error") || msg.contains("model_not_found") {
412        return ErrorCategory::NotFound;
413    }
414    // OpenRouter reports an unknown model as HTTP 400 with the body
415    // "<id> is not a valid model ID" — no status-code or typed-error signal
416    // that `classify_by_http_status` / the checks above can latch onto. Map
417    // the prose to NotFound so it lines up with Cerebras's 404 path (and with
418    // `errors::is_model_unavailable`'s reason taxonomy).
419    if lower.contains("is not a valid model id") || lower.contains("invalid model id") {
420        return ErrorCategory::NotFound;
421    }
422    if msg.contains("circuit_open") {
423        return ErrorCategory::CircuitOpen;
424    }
425    // Network-level transient patterns (pre-HTTP-status, pre-provider-framing).
426    if lower.contains("connection reset")
427        || lower.contains("connection refused")
428        || lower.contains("connection closed")
429        || lower.contains("broken pipe")
430        || lower.contains("dns error")
431        || lower.contains("stream error")
432        || lower.contains("unexpected eof")
433    {
434        return ErrorCategory::TransientNetwork;
435    }
436    ErrorCategory::Generic
437}
438
439/// Classify errors by HTTP status code if one appears in the message.
440/// This is the most reliable classification method since status codes
441/// are standardized (RFC 9110) and unambiguous.
442fn classify_by_http_status(msg: &str) -> Option<ErrorCategory> {
443    // Extract 3-digit HTTP status codes from common patterns:
444    // "HTTP 429", "status 429", "429 Too Many", "error: 401"
445    for code in extract_http_status_codes(msg) {
446        return Some(match code {
447            401 | 403 => ErrorCategory::Auth,
448            404 | 410 => ErrorCategory::NotFound,
449            408 | 504 | 522 | 524 => ErrorCategory::Timeout,
450            429 => ErrorCategory::RateLimit,
451            503 | 529 => ErrorCategory::Overloaded,
452            500 | 502 => ErrorCategory::ServerError,
453            _ => continue,
454        });
455    }
456    None
457}
458
459/// Extract plausible HTTP status codes from an error message.
460fn extract_http_status_codes(msg: &str) -> Vec<u16> {
461    let mut codes = Vec::new();
462    let bytes = msg.as_bytes();
463    for i in 0..bytes.len().saturating_sub(2) {
464        // Look for 3-digit sequences in the 100-599 range
465        if bytes[i].is_ascii_digit()
466            && bytes[i + 1].is_ascii_digit()
467            && bytes[i + 2].is_ascii_digit()
468        {
469            // Ensure it's not part of a longer number
470            let before_ok = i == 0 || !bytes[i - 1].is_ascii_digit();
471            let after_ok = i + 3 >= bytes.len() || !bytes[i + 3].is_ascii_digit();
472            if before_ok && after_ok {
473                if let Ok(code) = msg[i..i + 3].parse::<u16>() {
474                    if (400..=599).contains(&code) {
475                        codes.push(code);
476                    }
477                }
478            }
479        }
480    }
481    codes
482}
483
484impl std::fmt::Display for VmError {
485    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
486        match self {
487            VmError::StackUnderflow => write!(f, "Stack underflow"),
488            VmError::StackOverflow => write!(f, "Stack overflow: too many nested calls"),
489            VmError::UndefinedVariable(n) => write!(f, "Undefined variable: {n}"),
490            VmError::UndefinedBuiltin(n) => write!(f, "Undefined builtin: {n}"),
491            VmError::ImmutableAssignment(n) => {
492                write!(f, "Cannot assign to immutable binding: {n}")
493            }
494            VmError::TypeError(msg) => write!(f, "Type error: {msg}"),
495            VmError::Runtime(msg) => write!(f, "Runtime error: {msg}"),
496            VmError::DivisionByZero => write!(f, "Division by zero"),
497            VmError::ExecutionDeadlineExceeded => write!(f, "Execution deadline exceeded"),
498            VmError::ProcessExit(code) => write!(f, "Process exit requested: {code}"),
499            VmError::AbandonedExecution => write!(
500                f,
501                "Execution future was abandoned; discard this VM and reset its exclusively owned execution context"
502            ),
503            VmError::Thrown(v) => write!(f, "Thrown: {}", v.display()),
504            VmError::CategorizedError { message, category } => {
505                write!(f, "Error [{}]: {}", category.as_str(), message)
506            }
507            VmError::DaemonQueueFull {
508                daemon_id,
509                capacity,
510            } => write!(
511                f,
512                "Daemon queue full: daemon '{daemon_id}' reached its event_queue_capacity of {capacity}"
513            ),
514            VmError::Deadlock(err) => match err.diagnostic {
515                DeadlockDiagnostic::SelfDeadlock => write!(
516                    f,
517                    "{}: deadlock detected: {} ({} '{}') — this wait can never complete and would block forever",
518                    err.diagnostic.code(),
519                    err.detail,
520                    err.kind,
521                    err.key
522                ),
523                DeadlockDiagnostic::WaitForGraph => write!(
524                    f,
525                    "{}: wait-for deadlock detected: {} ({} '{}') — no active task can make progress",
526                    err.diagnostic.code(),
527                    err.detail,
528                    err.kind,
529                    err.key
530                ),
531            },
532            VmError::Return(_) => write!(f, "Return from function"),
533            VmError::InvalidInstruction(op) => write!(f, "Invalid instruction: 0x{op:02x}"),
534            VmError::ArityMismatch(err) => {
535                let arg_word = match err.expected {
536                    ArityExpect::Exact(1) | ArityExpect::AtLeast(1) => "argument",
537                    _ => "arguments",
538                };
539                write!(
540                    f,
541                    "Arity mismatch: '{}' expects {} {}, got {}{}",
542                    err.callee,
543                    err.expected,
544                    arg_word,
545                    err.got,
546                    fmt_span_suffix(&err.span)
547                )
548            }
549            VmError::ArgTypeMismatch(err) => {
550                write!(
551                    f,
552                    "Type error: '{}' parameter `{}` expects {}, got {}{}",
553                    err.callee,
554                    err.param,
555                    err.expected,
556                    err.got,
557                    fmt_span_suffix(&err.span)
558                )
559            }
560        }
561    }
562}
563
564fn fmt_span_suffix(span: &Option<Span>) -> String {
565    match span {
566        Some(s) => format!(" (at byte {}..{})", s.start, s.end),
567        None => String::new(),
568    }
569}
570
571impl std::error::Error for VmError {}
572
573#[cfg(test)]
574mod tests {
575    use super::*;
576
577    /// Every category. `as_str` already matches exhaustively, so a new
578    /// variant fails to compile until it is spelled there; this list is what
579    /// the round-trip and documentation guards below enumerate.
580    const ALL_CATEGORIES: &[ErrorCategory] = &[
581        ErrorCategory::Timeout,
582        ErrorCategory::Auth,
583        ErrorCategory::RateLimit,
584        ErrorCategory::Overloaded,
585        ErrorCategory::ServerError,
586        ErrorCategory::TransientNetwork,
587        ErrorCategory::SchemaValidation,
588        ErrorCategory::SchemaStreamAborted,
589        ErrorCategory::ToolError,
590        ErrorCategory::ToolRejected,
591        ErrorCategory::EgressBlocked,
592        ErrorCategory::Cancelled,
593        ErrorCategory::ChannelClosed,
594        ErrorCategory::NotFound,
595        ErrorCategory::CircuitOpen,
596        ErrorCategory::BudgetExceeded,
597        ErrorCategory::Internal,
598        ErrorCategory::Generic,
599    ];
600
601    /// A new variant must be added to `ALL_CATEGORIES`, or the guards below
602    /// silently stop covering it. This match is the tripwire: it fails to
603    /// compile until the variant is named, and the arm points at the list.
604    #[test]
605    fn all_categories_is_exhaustive() {
606        for category in ALL_CATEGORIES {
607            match category {
608                ErrorCategory::Timeout
609                | ErrorCategory::Auth
610                | ErrorCategory::RateLimit
611                | ErrorCategory::Overloaded
612                | ErrorCategory::ServerError
613                | ErrorCategory::TransientNetwork
614                | ErrorCategory::SchemaValidation
615                | ErrorCategory::SchemaStreamAborted
616                | ErrorCategory::ToolError
617                | ErrorCategory::ToolRejected
618                | ErrorCategory::EgressBlocked
619                | ErrorCategory::Cancelled
620                | ErrorCategory::ChannelClosed
621                | ErrorCategory::NotFound
622                | ErrorCategory::CircuitOpen
623                | ErrorCategory::BudgetExceeded
624                | ErrorCategory::Internal
625                | ErrorCategory::Generic => {}
626            }
627        }
628        assert_eq!(
629            ALL_CATEGORIES.len(),
630            18,
631            "a category was added or removed — update ALL_CATEGORIES and the \
632             `Error categories` table in docs/src/builtins.md"
633        );
634    }
635
636    #[test]
637    fn every_category_round_trips_through_parse() {
638        for category in ALL_CATEGORIES {
639            assert_eq!(
640                &ErrorCategory::parse(category.as_str()),
641                category,
642                "`{}` does not round-trip — `parse` is missing an arm, so a \
643                 host handing this category back to Harn silently gets \
644                 `generic`",
645                category.as_str()
646            );
647        }
648    }
649
650    /// Scripts branch on these strings, so an undocumented category is a
651    /// caller writing a match that cannot handle a value the runtime emits.
652    /// `error_category()` used to advertise 10 of the 18 — a dead-port probe
653    /// returning `transient_network` fell outside its own documented list.
654    #[test]
655    fn every_category_is_documented_in_builtins_md() {
656        let path = concat!(env!("CARGO_MANIFEST_DIR"), "/../../docs/src/builtins.md");
657        let doc =
658            std::fs::read_to_string(path).unwrap_or_else(|err| panic!("cannot read {path}: {err}"));
659        let table = doc
660            .split_once("### Error categories")
661            .unwrap_or_else(|| {
662                panic!("docs/src/builtins.md lost its `### Error categories` section")
663            })
664            .1;
665        let table = table.split_once("\n## ").map_or(table, |(head, _)| head);
666        for category in ALL_CATEGORIES {
667            let row = format!("| `{}` |", category.as_str());
668            assert!(
669                table.contains(&row),
670                "`{}` is missing from the `Error categories` table in \
671                 docs/src/builtins.md",
672                category.as_str()
673            );
674        }
675    }
676
677    #[test]
678    fn classifies_cancelled_messages() {
679        assert_eq!(
680            classify_error_message("Bridge: operation cancelled"),
681            ErrorCategory::Cancelled
682        );
683        assert_eq!(
684            classify_error_message("operation canceled by host"),
685            ErrorCategory::Cancelled
686        );
687    }
688
689    #[test]
690    fn classifies_undefined_builtin_as_internal() {
691        // Structured variant (dispatch table miss / uninstalled builtin).
692        assert_eq!(
693            error_to_category(&VmError::UndefinedBuiltin("__host_agent_foo".into())),
694            ErrorCategory::Internal
695        );
696        // Corrupt bytecode / compiler-VM opcode drift.
697        assert_eq!(
698            error_to_category(&VmError::InvalidInstruction(200)),
699            ErrorCategory::Internal
700        );
701        // Stringly form: some call sites build a `Runtime("Undefined builtin: …")`
702        // message instead of the structured variant — both must classify the same.
703        assert_eq!(
704            error_to_category(&VmError::Runtime(
705                "Undefined builtin: __host_agent_foo (did you mean `bar`?)".into()
706            )),
707            ErrorCategory::Internal
708        );
709        assert_eq!(
710            classify_error_message("Undefined builtin: __host_agent_foo"),
711            ErrorCategory::Internal
712        );
713        // Internal errors are never treated as transient/retryable.
714        assert!(!ErrorCategory::Internal.is_transient());
715        assert!(ErrorCategory::Internal.is_internal());
716        // Round-trips through the string form the agent loop compares against.
717        assert_eq!(ErrorCategory::Internal.as_str(), "internal");
718        assert_eq!(ErrorCategory::parse("internal"), ErrorCategory::Internal);
719    }
720
721    #[test]
722    fn classifies_openrouter_invalid_model_id_as_not_found() {
723        // OpenRouter reports an unknown model as HTTP 400 + prose. The 400 is
724        // not classified by status, so the prose substring must lift it to
725        // NotFound to match Cerebras's 404 path.
726        assert_eq!(
727            classify_error_message(
728                "openrouter API error: qwen/qwen3-coder-bogus is not a valid model ID"
729            ),
730            ErrorCategory::NotFound
731        );
732        assert_eq!(
733            classify_error_message("invalid model id supplied"),
734            ErrorCategory::NotFound
735        );
736    }
737
738    #[test]
739    fn categorized_error_lowers_to_structured_dict() {
740        // A caught `CategorizedError` must surface as a `{category, message}`
741        // dict so `.harn` consumers branch on the typed category instead of
742        // substring-matching the rendered prose (issue #4420).
743        let err = categorized_error(
744            "sandbox violation: /etc/passwd",
745            ErrorCategory::ToolRejected,
746        );
747        let VmValue::Dict(dict) = err.thrown_value() else {
748            panic!(
749                "categorized error must lower to a dict, got {:?}",
750                err.thrown_value()
751            );
752        };
753        assert_eq!(
754            dict.get("category").map(|v| v.display()).as_deref(),
755            Some("tool_rejected"),
756        );
757        assert_eq!(
758            dict.get("message").map(|v| v.display()).as_deref(),
759            Some("sandbox violation: /etc/passwd"),
760        );
761        // The key is the canonical, exhaustively-matched `ErrorCategory::as_str`
762        // contract — not the Display prose. A stringified catch still renders the
763        // message + category (so generic catch-and-log stays sensible).
764        let rendered = categorized_error("boom", ErrorCategory::Cancelled)
765            .thrown_value()
766            .display();
767        assert!(rendered.contains("cancelled"), "rendered dict: {rendered}");
768        assert!(rendered.contains("boom"), "rendered dict: {rendered}");
769    }
770
771    #[test]
772    fn thrown_value_passes_structured_thrown_through_unchanged() {
773        // A user `throw` of a structured value keeps its exact shape — the
774        // lowering seam must not stringify or re-wrap it.
775        let original = VmValue::dict(std::collections::BTreeMap::from([(
776            "code".to_string(),
777            VmValue::Int(7),
778        )]));
779        let VmValue::Dict(dict) = VmError::Thrown(original).thrown_value() else {
780            panic!("thrown dict must pass through as a dict");
781        };
782        assert!(matches!(dict.get("code"), Some(VmValue::Int(7))));
783    }
784
785    #[test]
786    fn deadlock_renders_with_stable_code() {
787        let err = VmError::Deadlock(Box::new(DeadlockError::self_deadlock(
788            "mutex",
789            "__default__",
790            "re-entrant acquire",
791        )));
792        assert!(
793            err.to_string().starts_with("HARN-ORC-011"),
794            "deadlock Display must carry the stable code: {err}"
795        );
796    }
797
798    #[test]
799    fn deadlock_maps_to_generic_category() {
800        let err = VmError::Deadlock(Box::new(DeadlockError::self_deadlock(
801            "task",
802            "task_1",
803            "self-join",
804        )));
805        let category = error_to_category(&err);
806        assert_eq!(category, ErrorCategory::Generic);
807        assert!(
808            !category.is_transient(),
809            "a deadlock must not be treated as a retryable transient error"
810        );
811    }
812}
harn_vm/value/error.rs

harn_vm/value/
error.rs
