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/// A value did not satisfy the type declared on the `let` / `const` it was
101/// being bound to.
102///
103/// Deliberately separate from [`ArgTypeMismatchError`]: both report a declared
104/// type rejecting a value, but the reader needs to know which binding site
105/// failed, and "parameter" is the wrong noun for half of them.
106#[derive(Debug, Clone)]
107pub struct BindingTypeMismatchError {
108    /// The declared name, or the rendered pattern for a destructuring binding.
109    pub binding: String,
110    pub expected: String,
111    pub got: &'static str,
112    pub span: Option<Span>,
113}
114
115#[derive(Debug, Clone)]
116pub enum VmError {
117    StackUnderflow,
118    StackOverflow,
119    UndefinedVariable(String),
120    UndefinedBuiltin(String),
121    ImmutableAssignment(String),
122    TypeError(String),
123    Runtime(String),
124    DivisionByZero,
125    /// A host-imposed deadline expired while executing the VM. Unlike a Harn
126    /// `deadline` block, this control-plane stop cannot be caught by user code.
127    ExecutionDeadlineExceeded,
128    /// A Harn program requested that its embedding process terminate with this
129    /// status code. Like a host deadline, this is control flow rather than a
130    /// catchable Harn error; the embedding boundary owns the final cleanup and
131    /// process exit.
132    ProcessExit(i32),
133    /// A host dropped a polled top-level execution future. Interpreter state is
134    /// intentionally not resumed after arbitrary async cancellation. Discard
135    /// this VM; see [`crate::Vm::execute_with_timeout`] for ambient-state
136    /// cleanup requirements.
137    AbandonedExecution,
138    /// A stable MCP handler needs client input before it can complete. The MCP
139    /// server boundary converts this uncatchable control signal into an
140    /// `input_required` result and re-enters the handler on retry.
141    McpInputRequired(Box<crate::mcp_input::McpInputRequired>),
142    Thrown(VmValue),
143    /// Thrown with error category for structured error handling.
144    CategorizedError {
145        message: String,
146        category: ErrorCategory,
147    },
148    /// A provider stream failed before its protocol supplied a terminal
149    /// sentinel. Carries transport phase and deadline provenance structurally
150    /// so retry, transcript, and host projections never need to reclassify
151    /// rendered prose.
152    ProviderStreamFailure(Box<ProviderStreamFailure>),
153    /// A streamed model response became impossible under its output schema.
154    /// Carries the validator's closed reason kind and exact detail so retries,
155    /// caught values, and transcript receipts never rebuild cause data from
156    /// the human-readable message.
157    SchemaStreamAbort(Box<SchemaStreamAbort>),
158    DaemonQueueFull {
159        daemon_id: String,
160        capacity: usize,
161    },
162    /// A deterministic, provably-unresolvable self-deadlock caught before the
163    /// VM would block forever (Rust's borrow checker prevents data races but
164    /// not deadlocks; this is the Go-runtime "all goroutines asleep" analogue
165    /// for the cases we can prove). Boxed — like [`VmError::ArityMismatch`] —
166    /// so the rare three-`String` payload doesn't enlarge `VmError` on the
167    /// pervasive `Result<VmValue, VmError>` hot path. Carries `HARN-ORC-011`.
168    Deadlock(Box<DeadlockError>),
169    Return(VmValue),
170    InvalidInstruction(u8),
171    /// Wrong number of arguments at a call site. Distinct from
172    /// [`VmError::TypeError`] so the runtime can match-and-recover (and
173    /// so error UX renders `expected 2..=3 got 1` consistently).
174    ArityMismatch(Box<ArityMismatchError>),
175    /// Argument value did not satisfy the declared parameter type.
176    /// `expected` is a pretty-printed type expression; `got` is the value's
177    /// runtime type name (`VmValue::type_name`). Used for both
178    /// user-defined function parameters (with declared types) and
179    /// registry-known builtin parameters.
180    ArgTypeMismatch(Box<ArgTypeMismatchError>),
181    /// Initializer value did not satisfy the type declared on its binding.
182    /// The binding-site counterpart of [`VmError::ArgTypeMismatch`]; a
183    /// declared type is checked wherever it is written.
184    BindingTypeMismatch(Box<BindingTypeMismatchError>),
185}
186
187impl VmError {
188    /// Whether this error is VM control flow that user `catch` blocks and
189    /// error-as-data combinators must propagate unchanged.
190    pub fn is_uncatchable_control_flow(&self) -> bool {
191        matches!(
192            self,
193            Self::ExecutionDeadlineExceeded | Self::ProcessExit(_) | Self::McpInputRequired(_)
194        )
195    }
196
197    /// The requested host-process exit status, when this is an explicit Harn
198    /// `exit(code)` control signal.
199    pub fn process_exit_code(&self) -> Option<i32> {
200        match self {
201            Self::ProcessExit(code) => Some(*code),
202            _ => None,
203        }
204    }
205
206    /// The `VmValue` a `catch` binding (or a `parallel settle` result) observes
207    /// for this error: the raw thrown value for [`VmError::Thrown`] (so a
208    /// structured error — e.g. a `{category, message}` dict from `throw_error` —
209    /// keeps its shape and category), a structured `{category, message}` dict for
210    /// [`VmError::CategorizedError`] (so consumers branch on the typed category
211    /// rather than substring-matching rendered prose), otherwise the rendered
212    /// message.
213    ///
214    /// Single source of truth for VM-error-to-value lowering so every seam that
215    /// surfaces a caught error to Harn (`try`/`catch` via `handle_error`,
216    /// `parallel settle`) exposes identical, structure-preserving values. Before
217    /// this was shared, `parallel settle` stringified errors via `to_string()`,
218    /// so a categorized error thrown in a settle branch lost its category (a
219    /// `cancelled`/`internal` fault that must propagate looked `generic`).
220    ///
221    /// The `category` key uses [`ErrorCategory::as_str`] — a canonical,
222    /// exhaustively-matched snake_case contract — so a new variant added there
223    /// is compiler-forced to name its key. `message` preserves the original
224    /// rendered text, so a `catch` that stringifies the caught value still reads
225    /// sensibly (the dict renders both fields).
226    pub fn thrown_value(&self) -> VmValue {
227        match self {
228            VmError::Thrown(v) => v.clone(),
229            VmError::CategorizedError { message, category } => {
230                let mut dict = std::collections::BTreeMap::new();
231                dict.put_str("category", category.as_str());
232                dict.put_str("message", message);
233                VmValue::dict(dict)
234            }
235            VmError::ProviderStreamFailure(failure) => failure.thrown_value(),
236            VmError::SchemaStreamAbort(abort) => abort.thrown_value(),
237            other => VmValue::String(arcstr::ArcStr::from(other.to_string())),
238        }
239    }
240
241    pub fn provider_stream_failure(&self) -> Option<&ProviderStreamFailure> {
242        match self {
243            Self::ProviderStreamFailure(failure) => Some(failure),
244            _ => None,
245        }
246    }
247
248    pub fn schema_stream_abort(&self) -> Option<&SchemaStreamAbort> {
249        match self {
250            Self::SchemaStreamAbort(abort) => Some(abort),
251            _ => None,
252        }
253    }
254}
255
256#[derive(Debug, Clone, Copy, PartialEq, Eq)]
257pub enum ProviderStreamPhase {
258    AwaitingFirstChunk,
259    Streaming,
260}
261
262impl ProviderStreamPhase {
263    pub fn as_str(self) -> &'static str {
264        match self {
265            Self::AwaitingFirstChunk => "awaiting_first_chunk",
266            Self::Streaming => "streaming",
267        }
268    }
269}
270
271#[derive(Debug, Clone, Copy, PartialEq, Eq)]
272pub enum ProviderStreamFailureReason {
273    Read,
274    PrematureEof,
275    Deadline,
276}
277
278impl ProviderStreamFailureReason {
279    pub fn as_str(self) -> &'static str {
280        match self {
281            Self::Read => "read",
282            Self::PrematureEof => "premature_eof",
283            Self::Deadline => "deadline",
284        }
285    }
286}
287
288#[derive(Debug, Clone, Copy, PartialEq, Eq)]
289pub enum ProviderStreamDeadline {
290    Total,
291    FirstChunk,
292    Idle,
293}
294
295impl ProviderStreamDeadline {
296    pub fn as_str(self) -> &'static str {
297        match self {
298            Self::Total => "total",
299            Self::FirstChunk => "first_chunk",
300            Self::Idle => "idle",
301        }
302    }
303}
304
305#[derive(Debug, Clone, PartialEq, Eq)]
306pub struct ProviderStreamFailure {
307    pub provider: String,
308    pub phase: ProviderStreamPhase,
309    pub reason: ProviderStreamFailureReason,
310    pub deadline: Option<ProviderStreamDeadline>,
311    pub partial: bool,
312    pub detail: String,
313}
314
315impl ProviderStreamFailure {
316    pub fn category(&self) -> ErrorCategory {
317        if self.deadline.is_some() {
318            ErrorCategory::Timeout
319        } else {
320            ErrorCategory::TransientNetwork
321        }
322    }
323
324    fn thrown_value(&self) -> VmValue {
325        let mut dict = std::collections::BTreeMap::new();
326        dict.put_str("category", self.category().as_str());
327        dict.put_str("message", self.to_string());
328        dict.put_str("source", "provider_stream");
329        dict.put_str("phase", self.phase.as_str());
330        dict.put_str("reason", self.reason.as_str());
331        dict.insert(
332            "deadline".to_string(),
333            self.deadline
334                .map(|deadline| VmValue::String(arcstr::ArcStr::from(deadline.as_str())))
335                .unwrap_or(VmValue::Nil),
336        );
337        dict.insert("partial".to_string(), VmValue::Bool(self.partial));
338        VmValue::dict(dict)
339    }
340}
341
342impl std::fmt::Display for ProviderStreamFailure {
343    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
344        write!(
345            f,
346            "{} provider stream failure (phase={}, reason={}",
347            self.provider,
348            self.phase.as_str(),
349            self.reason.as_str()
350        )?;
351        if let Some(deadline) = self.deadline {
352            write!(f, ", deadline={}", deadline.as_str())?;
353        }
354        write!(f, ", partial={}): {}", self.partial, self.detail)
355    }
356}
357
358/// Stable cause classes for JSON and schema validation failures.
359#[derive(Debug, Clone, Copy, PartialEq, Eq)]
360pub enum SchemaValidationReasonKind {
361    InvalidJson,
362    InvalidSchema,
363    WrongType,
364    MissingRequired,
365    UnexpectedProperty,
366    MaxLength,
367    MinLength,
368    ConstraintViolation,
369}
370
371impl SchemaValidationReasonKind {
372    pub fn as_str(self) -> &'static str {
373        match self {
374            Self::InvalidJson => "invalid_json",
375            Self::InvalidSchema => "invalid_schema",
376            Self::WrongType => "wrong_type",
377            Self::MissingRequired => "missing_required",
378            Self::UnexpectedProperty => "unexpected_property",
379            Self::MaxLength => "max_length",
380            Self::MinLength => "min_length",
381            Self::ConstraintViolation => "constraint_violation",
382        }
383    }
384}
385
386/// Typed cause retained when incremental schema validation aborts a provider
387/// stream.
388#[derive(Debug, Clone, PartialEq, Eq)]
389pub struct SchemaStreamAbort {
390    pub provider: String,
391    pub model: String,
392    pub reason_kind: SchemaValidationReasonKind,
393    pub reason: String,
394    pub path: String,
395    pub chunks_consumed: usize,
396}
397
398impl SchemaStreamAbort {
399    pub fn category(&self) -> ErrorCategory {
400        ErrorCategory::SchemaStreamAborted
401    }
402
403    fn thrown_value(&self) -> VmValue {
404        let mut cause = std::collections::BTreeMap::new();
405        cause.put_str("kind", self.reason_kind.as_str());
406        cause.put_str("detail", self.reason.as_str());
407        cause.put_str("path", self.path.as_str());
408        cause.insert(
409            "chunks_consumed".to_string(),
410            VmValue::Int(self.chunks_consumed as i64),
411        );
412        cause.put_str("provider", self.provider.as_str());
413        cause.put_str("model", self.model.as_str());
414
415        let mut dict = std::collections::BTreeMap::new();
416        dict.put_str("category", self.category().as_str());
417        dict.put_str("message", self.to_string());
418        dict.put_str("source", "schema_stream");
419        dict.insert("schema_failure".to_string(), VmValue::dict(cause));
420        VmValue::dict(dict)
421    }
422}
423
424impl std::fmt::Display for SchemaStreamAbort {
425    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
426        write!(
427            f,
428            "schema_stream_aborted at {}: {} (provider={} model={} chunks_consumed={})",
429            self.path, self.reason, self.provider, self.model, self.chunks_consumed
430        )
431    }
432}
433
434/// Error categories for structured error handling in agent orchestration.
435#[derive(Debug, Clone, PartialEq, Eq)]
436pub enum ErrorCategory {
437    /// Network/connection timeout
438    Timeout,
439    /// Authentication/authorization failure
440    Auth,
441    /// Provider rejected the request. Correct it before retrying.
442    InvalidRequest,
443    /// Rate limit exceeded (HTTP 429 / quota)
444    RateLimit,
445    /// Upstream provider is overloaded (HTTP 503 / 529).
446    /// Distinct from RateLimit: the client hasn't exceeded a quota — the
447    /// provider is shedding load and will recover on its own.
448    Overloaded,
449    /// Provider-side 5xx error (500, 502) that isn't specifically overload.
450    ServerError,
451    /// Network-level transient failure (connection reset, DNS hiccup,
452    /// partial stream) — retryable but not provider-status-coded.
453    TransientNetwork,
454    /// A shared local resource is temporarily unavailable, such as a
455    /// contended database write lock.
456    ResourceBusy,
457    /// A persistent store was written by a newer incompatible schema owner.
458    /// Retrying cannot help; the caller must upgrade or deliberately degrade.
459    SchemaIncompatible,
460    /// LLM output failed schema validation. Retryable via `schema_retries`.
461    SchemaValidation,
462    /// LLM streaming response was aborted mid-stream because the partial
463    /// JSON content could not conceivably satisfy `output_schema`. Surfaced
464    /// by `llm_call` when `schema_stream_abort` is on (the default for
465    /// schema-bearing calls). Consumes one `schema_retries` budget slot;
466    /// the retry replays the prompt with a corrective nudge that cites
467    /// the abort path + reason.
468    SchemaStreamAborted,
469    /// Tool execution failure
470    ToolError,
471    /// Tool was rejected by the host (not permitted / not in allowlist)
472    ToolRejected,
473    /// Outbound network egress was blocked by policy.
474    EgressBlocked,
475    /// Operation was cancelled
476    Cancelled,
477    /// Channel was closed before the operation could complete.
478    ChannelClosed,
479    /// Resource not found
480    NotFound,
481    /// Circuit breaker is open
482    CircuitOpen,
483    /// LLM cost or token budget would be exceeded
484    BudgetExceeded,
485    /// An internal engine/wiring bug — an undefined builtin, corrupt bytecode,
486    /// or another VM invariant violation that no amount of retrying or model
487    /// reasoning can fix. Distinct from `Generic` so callers (notably the agent
488    /// loop) can re-raise it loudly instead of folding it into a tool-error
489    /// observation and marching on to a `done` status. This is the category
490    /// that keeps a mis-wired builtin (e.g. a `#[harn_builtin]` def missing
491    /// from its install array) from shipping silently inert.
492    Internal,
493    /// A host environment / infrastructure problem that is not the workload's
494    /// code defect: a required developer-toolchain root or cache lies outside
495    /// the sandbox profile, a needed system binary is missing, or another
496    /// machine-provisioning gap. Distinct from `ToolRejected` (the host
497    /// deliberately refused an action) and `Internal` (an engine bug): the fix
498    /// is to widen the sandbox/config or provision the host, not to change the
499    /// agent's code. Callers (and embedders) branch on this to avoid blaming
500    /// the model for an environment gap.
501    Environment,
502    /// Generic/unclassified error
503    Generic,
504}
505
506impl ErrorCategory {
507    /// Every category, in declaration order.
508    ///
509    /// Sibling taxonomies (`ToolCallErrorCategory::ALL`,
510    /// `AgentTerminalKind::ALL`) already publish theirs, and code that has to
511    /// decide something for EVERY category — a wire projection, a docs table,
512    /// a round-trip guard — needs to enumerate them. While this list lived in
513    /// one module's test scope, the tool-call wire projection could not consult
514    /// it, and a category with no decided wire bucket went unnoticed (#5537).
515    pub const ALL: [Self; 22] = [
516        Self::Timeout,
517        Self::Auth,
518        Self::InvalidRequest,
519        Self::RateLimit,
520        Self::Overloaded,
521        Self::ServerError,
522        Self::TransientNetwork,
523        Self::ResourceBusy,
524        Self::SchemaIncompatible,
525        Self::SchemaValidation,
526        Self::SchemaStreamAborted,
527        Self::ToolError,
528        Self::ToolRejected,
529        Self::EgressBlocked,
530        Self::Cancelled,
531        Self::ChannelClosed,
532        Self::NotFound,
533        Self::CircuitOpen,
534        Self::BudgetExceeded,
535        Self::Internal,
536        Self::Environment,
537        Self::Generic,
538    ];
539
540    pub fn as_str(&self) -> &'static str {
541        match self {
542            ErrorCategory::Timeout => "timeout",
543            ErrorCategory::Auth => "auth",
544            ErrorCategory::InvalidRequest => "invalid_request",
545            ErrorCategory::RateLimit => "rate_limit",
546            ErrorCategory::Overloaded => "overloaded",
547            ErrorCategory::ServerError => "server_error",
548            ErrorCategory::TransientNetwork => "transient_network",
549            ErrorCategory::ResourceBusy => "resource_busy",
550            ErrorCategory::SchemaIncompatible => "schema_incompatible",
551            ErrorCategory::SchemaValidation => "schema_validation",
552            ErrorCategory::SchemaStreamAborted => "schema_stream_aborted",
553            ErrorCategory::ToolError => "tool_error",
554            ErrorCategory::ToolRejected => "tool_rejected",
555            ErrorCategory::EgressBlocked => "egress_blocked",
556            ErrorCategory::Cancelled => "cancelled",
557            ErrorCategory::ChannelClosed => "channel_closed",
558            ErrorCategory::NotFound => "not_found",
559            ErrorCategory::CircuitOpen => "circuit_open",
560            ErrorCategory::BudgetExceeded => "budget_exceeded",
561            ErrorCategory::Internal => "internal",
562            ErrorCategory::Environment => "environment",
563            ErrorCategory::Generic => "generic",
564        }
565    }
566
567    pub fn parse(s: &str) -> Self {
568        match s {
569            "timeout" => ErrorCategory::Timeout,
570            "auth" => ErrorCategory::Auth,
571            "invalid_request" => ErrorCategory::InvalidRequest,
572            "rate_limit" => ErrorCategory::RateLimit,
573            "overloaded" => ErrorCategory::Overloaded,
574            "server_error" => ErrorCategory::ServerError,
575            "transient_network" => ErrorCategory::TransientNetwork,
576            "resource_busy" => ErrorCategory::ResourceBusy,
577            "schema_incompatible" => ErrorCategory::SchemaIncompatible,
578            "schema_validation" => ErrorCategory::SchemaValidation,
579            "schema_stream_aborted" => ErrorCategory::SchemaStreamAborted,
580            "tool_error" => ErrorCategory::ToolError,
581            "tool_rejected" => ErrorCategory::ToolRejected,
582            "egress_blocked" => ErrorCategory::EgressBlocked,
583            "cancelled" => ErrorCategory::Cancelled,
584            "channel_closed" => ErrorCategory::ChannelClosed,
585            "not_found" => ErrorCategory::NotFound,
586            "circuit_open" => ErrorCategory::CircuitOpen,
587            "budget_exceeded" => ErrorCategory::BudgetExceeded,
588            "internal" => ErrorCategory::Internal,
589            "environment" => ErrorCategory::Environment,
590            _ => ErrorCategory::Generic,
591        }
592    }
593
594    /// Whether this category represents an internal engine/wiring bug that must
595    /// be surfaced rather than retried or swallowed as a recoverable failure.
596    pub fn is_internal(&self) -> bool {
597        matches!(self, ErrorCategory::Internal)
598    }
599
600    /// Whether an error of this category is worth retrying because the
601    /// underlying condition is transient. Agent loops consult this to decide
602    /// whether to back off and retry vs surface the error to the user.
603    pub fn is_transient(&self) -> bool {
604        matches!(
605            self,
606            ErrorCategory::Timeout
607                | ErrorCategory::RateLimit
608                | ErrorCategory::Overloaded
609                | ErrorCategory::ServerError
610                | ErrorCategory::TransientNetwork
611                | ErrorCategory::ResourceBusy
612        )
613    }
614}
615
616/// Create a categorized error conveniently.
617pub fn categorized_error(message: impl Into<String>, category: ErrorCategory) -> VmError {
618    VmError::CategorizedError {
619        message: message.into(),
620        category,
621    }
622}
623
624/// Extract error category from a VmError.
625///
626/// Classification priority:
627/// 1. Explicit CategorizedError variant (set by throw_error or internal code)
628/// 2. Thrown dict with a "category" field (user-created structured errors)
629/// 3. HTTP status code extraction (standard, unambiguous)
630/// 4. Deadline exceeded (VM-internal)
631/// 5. Fallback to Generic
632pub fn error_to_category(err: &VmError) -> ErrorCategory {
633    match err {
634        VmError::ExecutionDeadlineExceeded => ErrorCategory::Timeout,
635        // ProcessExit is uncatchable control flow rather than an agent-facing
636        // failure. Keep this fallback total for callers that classify an
637        // arbitrary VmError without treating the request as retryable.
638        VmError::ProcessExit(_) => ErrorCategory::Generic,
639        VmError::AbandonedExecution => ErrorCategory::Cancelled,
640        VmError::CategorizedError { category, .. } => category.clone(),
641        VmError::ProviderStreamFailure(failure) => failure.category(),
642        VmError::SchemaStreamAbort(abort) => abort.category(),
643        VmError::Thrown(VmValue::Dict(d)) => d
644            .get("category")
645            .map(|v| ErrorCategory::parse(&v.display()))
646            .unwrap_or(ErrorCategory::Generic),
647        VmError::Thrown(VmValue::String(s)) => classify_error_message(s),
648        VmError::Runtime(msg) => classify_error_message(msg),
649        // Engine/wiring bugs: an undefined builtin (declared but not installed,
650        // or a typo in stdlib/host code) or corrupt bytecode. No retry or model
651        // reasoning fixes these, so they get their own category the agent loop
652        // re-raises instead of swallowing.
653        VmError::UndefinedBuiltin(_) | VmError::InvalidInstruction(_) => ErrorCategory::Internal,
654        // A deadlock is permanently non-retryable and not provider-related —
655        // `Generic` is the correct "surface it, don't back off" bucket.
656        VmError::Deadlock(_) => ErrorCategory::Generic,
657        _ => ErrorCategory::Generic,
658    }
659}
660
661/// Classify an error message using HTTP status codes and well-known patterns.
662/// Prefers unambiguous signals (status codes) over substring heuristics.
663pub fn classify_error_message(msg: &str) -> ErrorCategory {
664    // 1. HTTP status codes — most reliable signal
665    if let Some(cat) = classify_by_http_status(msg) {
666        return cat;
667    }
668    // 2. Internal engine/wiring bug surfaced as a plain message. Some call
669    //    sites build `Runtime("Undefined builtin: …")` strings instead of the
670    //    structured `VmError::UndefinedBuiltin` variant; classify both the same
671    //    so the agent loop re-raises rather than swallows.
672    if msg.contains("Undefined builtin") {
673        return ErrorCategory::Internal;
674    }
675    // 3. Well-known error identifiers from major APIs
676    //    (Anthropic, OpenAI, and standard HTTP patterns)
677    let lower = msg.to_lowercase();
678    if lower.contains("cancelled") || lower.contains("canceled") {
679        return ErrorCategory::Cancelled;
680    }
681    if msg.contains("ChannelClosed") || lower.contains("channel closed") {
682        return ErrorCategory::ChannelClosed;
683    }
684    if msg.contains("Deadline exceeded") || msg.contains("context deadline exceeded") {
685        return ErrorCategory::Timeout;
686    }
687    if msg.contains("overloaded_error") {
688        // Anthropic overloaded_error surfaces as HTTP 529.
689        return ErrorCategory::Overloaded;
690    }
691    if msg.contains("api_error") {
692        // Anthropic catch-all server-side error.
693        return ErrorCategory::ServerError;
694    }
695    if msg.contains("insufficient_quota") || msg.contains("billing_hard_limit_reached") {
696        // OpenAI-specific quota error types.
697        return ErrorCategory::RateLimit;
698    }
699    if msg.contains("invalid_api_key") || msg.contains("authentication_error") {
700        return ErrorCategory::Auth;
701    }
702    if msg.contains("not_found_error") || msg.contains("model_not_found") {
703        return ErrorCategory::NotFound;
704    }
705    // OpenRouter reports an unknown model as HTTP 400 with the body
706    // "<id> is not a valid model ID" — no status-code or typed-error signal
707    // that `classify_by_http_status` / the checks above can latch onto. Map
708    // the prose to NotFound so it lines up with Cerebras's 404 path (and with
709    // `errors::is_model_unavailable`'s reason taxonomy).
710    if lower.contains("is not a valid model id") || lower.contains("invalid model id") {
711        return ErrorCategory::NotFound;
712    }
713    if msg.contains("circuit_open") {
714        return ErrorCategory::CircuitOpen;
715    }
716    // Network-level transient patterns (pre-HTTP-status, pre-provider-framing).
717    if lower.contains("connection reset")
718        || lower.contains("connection refused")
719        || lower.contains("connection closed")
720        || lower.contains("broken pipe")
721        || lower.contains("dns error")
722        || lower.contains("stream error")
723        || lower.contains("unexpected eof")
724    {
725        return ErrorCategory::TransientNetwork;
726    }
727    ErrorCategory::Generic
728}
729
730/// Classify errors by HTTP status code if one appears in the message.
731/// This is the most reliable classification method since status codes
732/// are standardized (RFC 9110) and unambiguous.
733fn classify_by_http_status(msg: &str) -> Option<ErrorCategory> {
734    // Extract 3-digit HTTP status codes from common patterns:
735    // "HTTP 429", "status 429", "429 Too Many", "error: 401"
736    for code in extract_http_status_codes(msg) {
737        if let Some(category) = error_category_for_http_status(code) {
738            return Some(category);
739        }
740    }
741    None
742}
743
744/// Map an explicit provider HTTP status to the shared runtime category.
745///
746/// Callers that already own a typed status should use this instead of
747/// formatting it into prose and asking [`classify_error_message`] to recover
748/// the same fact. Keeping the mapping here also prevents mock and live provider
749/// envelopes from drifting apart.
750pub(crate) fn error_category_for_http_status(status: u16) -> Option<ErrorCategory> {
751    match status {
752        401 | 403 => Some(ErrorCategory::Auth),
753        404 | 410 => Some(ErrorCategory::NotFound),
754        408 | 504 | 522 | 524 => Some(ErrorCategory::Timeout),
755        429 => Some(ErrorCategory::RateLimit),
756        503 | 529 => Some(ErrorCategory::Overloaded),
757        500 | 502 => Some(ErrorCategory::ServerError),
758        _ => None,
759    }
760}
761
762/// Extract plausible HTTP status codes from an error message.
763#[expect(
764    clippy::string_slice,
765    reason = "i..i + 3 spans bytes verified to be ASCII digits"
766)]
767fn extract_http_status_codes(msg: &str) -> Vec<u16> {
768    let mut codes = Vec::new();
769    let bytes = msg.as_bytes();
770    for i in 0..bytes.len().saturating_sub(2) {
771        // Look for 3-digit sequences in the 100-599 range
772        if bytes[i].is_ascii_digit()
773            && bytes[i + 1].is_ascii_digit()
774            && bytes[i + 2].is_ascii_digit()
775        {
776            // Ensure it's not part of a longer number
777            let before_ok = i == 0 || !bytes[i - 1].is_ascii_digit();
778            let after_ok = i + 3 >= bytes.len() || !bytes[i + 3].is_ascii_digit();
779            if before_ok && after_ok {
780                if let Ok(code) = msg[i..i + 3].parse::<u16>() {
781                    if (400..=599).contains(&code) {
782                        codes.push(code);
783                    }
784                }
785            }
786        }
787    }
788    codes
789}
790
791impl std::fmt::Display for VmError {
792    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
793        match self {
794            VmError::StackUnderflow => write!(f, "Stack underflow"),
795            VmError::StackOverflow => write!(f, "Stack overflow: too many nested calls"),
796            VmError::UndefinedVariable(n) => write!(f, "Undefined variable: {n}"),
797            VmError::UndefinedBuiltin(n) => write!(f, "Undefined builtin: {n}"),
798            VmError::ImmutableAssignment(n) => {
799                write!(f, "Cannot assign to immutable binding: {n}")
800            }
801            VmError::TypeError(msg) => write!(f, "Type error: {msg}"),
802            VmError::Runtime(msg) => write!(f, "Runtime error: {msg}"),
803            VmError::DivisionByZero => write!(f, "Division by zero"),
804            VmError::ExecutionDeadlineExceeded => write!(f, "Execution deadline exceeded"),
805            VmError::ProcessExit(code) => write!(f, "Process exit requested: {code}"),
806            VmError::AbandonedExecution => write!(
807                f,
808                "Execution future was abandoned; discard this VM and reset its exclusively owned execution context"
809            ),
810            VmError::McpInputRequired(_) => write!(f, "MCP client input required"),
811            VmError::Thrown(v) => write!(f, "Thrown: {}", v.display()),
812            VmError::CategorizedError { message, category } => {
813                write!(f, "Error [{}]: {}", category.as_str(), message)
814            }
815            VmError::ProviderStreamFailure(failure) => failure.fmt(f),
816            VmError::SchemaStreamAbort(abort) => abort.fmt(f),
817            VmError::DaemonQueueFull {
818                daemon_id,
819                capacity,
820            } => write!(
821                f,
822                "Daemon queue full: daemon '{daemon_id}' reached its event_queue_capacity of {capacity}"
823            ),
824            VmError::Deadlock(err) => match err.diagnostic {
825                DeadlockDiagnostic::SelfDeadlock => write!(
826                    f,
827                    "{}: deadlock detected: {} ({} '{}') — this wait can never complete and would block forever",
828                    err.diagnostic.code(),
829                    err.detail,
830                    err.kind,
831                    err.key
832                ),
833                DeadlockDiagnostic::WaitForGraph => write!(
834                    f,
835                    "{}: wait-for deadlock detected: {} ({} '{}') — no active task can make progress",
836                    err.diagnostic.code(),
837                    err.detail,
838                    err.kind,
839                    err.key
840                ),
841            },
842            VmError::Return(_) => write!(f, "Return from function"),
843            VmError::InvalidInstruction(op) => write!(f, "Invalid instruction: 0x{op:02x}"),
844            VmError::ArityMismatch(err) => {
845                let arg_word = match err.expected {
846                    ArityExpect::Exact(1) | ArityExpect::AtLeast(1) => "argument",
847                    _ => "arguments",
848                };
849                write!(
850                    f,
851                    "Arity mismatch: '{}' expects {} {}, got {}{}",
852                    err.callee,
853                    err.expected,
854                    arg_word,
855                    err.got,
856                    fmt_span_suffix(&err.span)
857                )
858            }
859            VmError::ArgTypeMismatch(err) => {
860                write!(
861                    f,
862                    "Type error: '{}' parameter `{}` expects {}, got {}{}",
863                    err.callee,
864                    err.param,
865                    err.expected,
866                    err.got,
867                    fmt_span_suffix(&err.span)
868                )
869            }
870            VmError::BindingTypeMismatch(err) => {
871                write!(
872                    f,
873                    "Type error: binding `{}` expects {}, got {}{}",
874                    err.binding,
875                    err.expected,
876                    err.got,
877                    fmt_span_suffix(&err.span)
878                )
879            }
880        }
881    }
882}
883
884fn fmt_span_suffix(span: &Option<Span>) -> String {
885    match span {
886        Some(s) => format!(" (at byte {}..{})", s.start, s.end),
887        None => String::new(),
888    }
889}
890
891impl std::error::Error for VmError {}
892
893#[cfg(test)]
894mod tests {
895    use super::*;
896
897    /// A new variant must be added to [`ErrorCategory::ALL`], or the guards below
898    /// silently stop covering it. This match is the tripwire: it fails to
899    /// compile until the variant is named, and the arm points at the list.
900    #[test]
901    fn all_categories_is_exhaustive() {
902        for category in &ErrorCategory::ALL {
903            match category {
904                ErrorCategory::Timeout
905                | ErrorCategory::Auth
906                | ErrorCategory::InvalidRequest
907                | ErrorCategory::RateLimit
908                | ErrorCategory::Overloaded
909                | ErrorCategory::ServerError
910                | ErrorCategory::TransientNetwork
911                | ErrorCategory::ResourceBusy
912                | ErrorCategory::SchemaIncompatible
913                | ErrorCategory::SchemaValidation
914                | ErrorCategory::SchemaStreamAborted
915                | ErrorCategory::ToolError
916                | ErrorCategory::ToolRejected
917                | ErrorCategory::EgressBlocked
918                | ErrorCategory::Cancelled
919                | ErrorCategory::ChannelClosed
920                | ErrorCategory::NotFound
921                | ErrorCategory::CircuitOpen
922                | ErrorCategory::BudgetExceeded
923                | ErrorCategory::Internal
924                | ErrorCategory::Environment
925                | ErrorCategory::Generic => {}
926            }
927        }
928        assert_eq!(
929            ErrorCategory::ALL.len(),
930            22,
931            "a category was added or removed — update `ErrorCategory::ALL` and the \
932             `Error categories` table in docs/src/builtins.md"
933        );
934    }
935
936    #[test]
937    fn every_category_round_trips_through_parse() {
938        for category in &ErrorCategory::ALL {
939            assert_eq!(
940                &ErrorCategory::parse(category.as_str()),
941                category,
942                "`{}` does not round-trip — `parse` is missing an arm, so a \
943                 host handing this category back to Harn silently gets \
944                 `generic`",
945                category.as_str()
946            );
947        }
948    }
949
950    /// Scripts branch on these strings, so an undocumented category is a
951    /// caller writing a match that cannot handle a value the runtime emits.
952    /// `error_category()` used to advertise 10 of the full category set — a dead-port probe
953    /// returning `transient_network` fell outside its own documented list.
954    #[test]
955    fn every_category_is_documented_in_builtins_md() {
956        let path = concat!(env!("CARGO_MANIFEST_DIR"), "/../../docs/src/builtins.md");
957        let doc =
958            std::fs::read_to_string(path).unwrap_or_else(|err| panic!("cannot read {path}: {err}"));
959        let table = doc
960            .split_once("### Error categories")
961            .unwrap_or_else(|| {
962                panic!("docs/src/builtins.md lost its `### Error categories` section")
963            })
964            .1;
965        let table = table.split_once("\n## ").map_or(table, |(head, _)| head);
966        for category in &ErrorCategory::ALL {
967            let row = format!("| `{}` |", category.as_str());
968            assert!(
969                table.contains(&row),
970                "`{}` is missing from the `Error categories` table in \
971                 docs/src/builtins.md",
972                category.as_str()
973            );
974        }
975    }
976
977    #[test]
978    fn classifies_cancelled_messages() {
979        assert_eq!(
980            classify_error_message("Bridge: operation cancelled"),
981            ErrorCategory::Cancelled
982        );
983        assert_eq!(
984            classify_error_message("operation canceled by host"),
985            ErrorCategory::Cancelled
986        );
987    }
988
989    #[test]
990    fn classifies_undefined_builtin_as_internal() {
991        // Structured variant (dispatch table miss / uninstalled builtin).
992        assert_eq!(
993            error_to_category(&VmError::UndefinedBuiltin("__host_agent_foo".into())),
994            ErrorCategory::Internal
995        );
996        // Corrupt bytecode / compiler-VM opcode drift.
997        assert_eq!(
998            error_to_category(&VmError::InvalidInstruction(200)),
999            ErrorCategory::Internal
1000        );
1001        // Stringly form: some call sites build a `Runtime("Undefined builtin: …")`
1002        // message instead of the structured variant — both must classify the same.
1003        assert_eq!(
1004            error_to_category(&VmError::Runtime(
1005                "Undefined builtin: __host_agent_foo (did you mean `bar`?)".into()
1006            )),
1007            ErrorCategory::Internal
1008        );
1009        assert_eq!(
1010            classify_error_message("Undefined builtin: __host_agent_foo"),
1011            ErrorCategory::Internal
1012        );
1013        // Internal errors are never treated as transient/retryable.
1014        assert!(!ErrorCategory::Internal.is_transient());
1015        assert!(ErrorCategory::Internal.is_internal());
1016        // Round-trips through the string form the agent loop compares against.
1017        assert_eq!(ErrorCategory::Internal.as_str(), "internal");
1018        assert_eq!(ErrorCategory::parse("internal"), ErrorCategory::Internal);
1019    }
1020
1021    #[test]
1022    fn classifies_openrouter_invalid_model_id_as_not_found() {
1023        // OpenRouter reports an unknown model as HTTP 400 + prose. The 400 is
1024        // not classified by status, so the prose substring must lift it to
1025        // NotFound to match Cerebras's 404 path.
1026        assert_eq!(
1027            classify_error_message(
1028                "openrouter API error: qwen/qwen3-coder-bogus is not a valid model ID"
1029            ),
1030            ErrorCategory::NotFound
1031        );
1032        assert_eq!(
1033            classify_error_message("invalid model id supplied"),
1034            ErrorCategory::NotFound
1035        );
1036    }
1037
1038    #[test]
1039    fn categorized_error_lowers_to_structured_dict() {
1040        // A caught `CategorizedError` must surface as a `{category, message}`
1041        // dict so `.harn` consumers branch on the typed category instead of
1042        // substring-matching the rendered prose (issue #4420).
1043        let err = categorized_error(
1044            "sandbox violation: /etc/passwd",
1045            ErrorCategory::ToolRejected,
1046        );
1047        let VmValue::Dict(dict) = err.thrown_value() else {
1048            panic!(
1049                "categorized error must lower to a dict, got {:?}",
1050                err.thrown_value()
1051            );
1052        };
1053        assert_eq!(
1054            dict.get("category").map(|v| v.display()).as_deref(),
1055            Some("tool_rejected"),
1056        );
1057        assert_eq!(
1058            dict.get("message").map(|v| v.display()).as_deref(),
1059            Some("sandbox violation: /etc/passwd"),
1060        );
1061        // The key is the canonical, exhaustively-matched `ErrorCategory::as_str`
1062        // contract — not the Display prose. A stringified catch still renders the
1063        // message + category (so generic catch-and-log stays sensible).
1064        let rendered = categorized_error("boom", ErrorCategory::Cancelled)
1065            .thrown_value()
1066            .display();
1067        assert!(rendered.contains("cancelled"), "rendered dict: {rendered}");
1068        assert!(rendered.contains("boom"), "rendered dict: {rendered}");
1069    }
1070
1071    #[test]
1072    fn thrown_value_passes_structured_thrown_through_unchanged() {
1073        // A user `throw` of a structured value keeps its exact shape — the
1074        // lowering seam must not stringify or re-wrap it.
1075        let original = VmValue::dict(std::collections::BTreeMap::from([(
1076            "code".to_string(),
1077            VmValue::Int(7),
1078        )]));
1079        let VmValue::Dict(dict) = VmError::Thrown(original).thrown_value() else {
1080            panic!("thrown dict must pass through as a dict");
1081        };
1082        assert!(matches!(dict.get("code"), Some(VmValue::Int(7))));
1083    }
1084
1085    #[test]
1086    fn deadlock_renders_with_stable_code() {
1087        let err = VmError::Deadlock(Box::new(DeadlockError::self_deadlock(
1088            "mutex",
1089            "__default__",
1090            "re-entrant acquire",
1091        )));
1092        assert!(
1093            err.to_string().starts_with("HARN-ORC-011"),
1094            "deadlock Display must carry the stable code: {err}"
1095        );
1096    }
1097
1098    #[test]
1099    fn deadlock_maps_to_generic_category() {
1100        let err = VmError::Deadlock(Box::new(DeadlockError::self_deadlock(
1101            "task",
1102            "task_1",
1103            "self-join",
1104        )));
1105        let category = error_to_category(&err);
1106        assert_eq!(category, ErrorCategory::Generic);
1107        assert!(
1108            !category.is_transient(),
1109            "a deadlock must not be treated as a retryable transient error"
1110        );
1111    }
1112}
harn_vm/value/error.rs

harn_vm/value/
error.rs
