rudb_common/
error.rs

1//! The error model.
2//!
3//! `spec/04-architecture.md` section 4.9 says errors are values, `Result` is everywhere, and no
4//! path reachable from user input panics. It also says every error carries a code, a message and
5//! optionally a span into the query text, that the codes are stable because clients switch on
6//! them, and that the messages match DuckDB's where a DuckDB message is what a test asserts on.
7//!
8//! This module is where all three of those obligations live.
9
10use std::fmt;
11
12/// The result type used everywhere in the workspace.
13pub type Result<T> = std::result::Result<T, Error>;
14
15/// A byte range into the query text.
16///
17/// Half open, so `start` is the first byte and `end` is one past the last, which is what slicing
18/// wants and what every editor protocol in existence expects. Byte offsets rather than character
19/// offsets because that is what the parser has and converting is the caller's problem, once, at
20/// the point where a human is going to read it.
21#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
22pub struct Span {
23    /// First byte of the span.
24    pub start: u32,
25    /// One past the last byte of the span.
26    pub end: u32,
27}
28
29impl Span {
30    /// A span over `start .. end`.
31    #[must_use]
32    pub const fn new(start: u32, end: u32) -> Self {
33        Self { start, end }
34    }
35
36    /// The number of bytes covered, which is zero for a span that points between two characters.
37    #[must_use]
38    pub const fn len(self) -> u32 {
39        self.end.saturating_sub(self.start)
40    }
41
42    /// Whether the span covers no bytes.
43    #[must_use]
44    pub const fn is_empty(self) -> bool {
45        self.len() == 0
46    }
47}
48
49/// What kind of thing went wrong.
50///
51/// These are stable and they are part of the public interface, because a client that retries on
52/// one class of failure and gives up on another has to be able to tell them apart without reading
53/// the message. Adding a variant is a compatible change and renaming one is not.
54#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
55#[non_exhaustive]
56pub enum ErrorCode {
57    /// The text is not SQL.
58    Parser,
59    /// The text is SQL and a value in it is not one the statement can take.
60    ///
61    /// DuckDB's own line between this and [`ErrorCode::Parser`] is not one anybody would draw
62    /// twice, and it is on the wire, so it is here: `SET threads=0` is a syntax error there and a
63    /// syntax error here.
64    Syntax,
65    /// The text is SQL and it does not mean anything, for example a column that is not in scope.
66    Binder,
67    /// A named object is missing, or one that should be missing is not.
68    Catalog,
69    /// A value will not convert to the type it is being asked for.
70    Conversion,
71    /// A value is outside what its type can hold.
72    OutOfRange,
73    /// An argument is wrong in a way that is not a type error, for example a negative length.
74    InvalidInput,
75    /// An allocation failed or a memory limit was reached. An error, never an abort.
76    OutOfMemory,
77    /// The filesystem, the network or the object store said no.
78    Io,
79    /// It is in the plan and it is not built yet.
80    NotImplemented,
81    /// A primary key, unique, not null or check constraint was violated.
82    Constraint,
83    /// A conflict, an abort, or a statement issued outside a transaction that needs one.
84    Transaction,
85    /// A setting cannot be applied in the current engine configuration.
86    Settings,
87    /// The query was cancelled. Cooperative, checked at morsel boundaries.
88    Interrupt,
89    /// A value is one the function refuses outright rather than one of the wrong type, for example
90    /// an empty list handed to `list_reduce` with nothing to start from.
91    ParameterNotAllowed,
92    /// Two types were asked to meet and cannot, for example two structs of different sizes.
93    MismatchType,
94    /// An invariant this code is responsible for does not hold. Always a bug here, never in the
95    /// query.
96    Internal,
97}
98
99impl ErrorCode {
100    /// The prefix DuckDB puts on a message with this code.
101    ///
102    /// Compatibility obligation from `spec/12-duckdb-compat.md` section 12.5: a great many tests
103    /// in the wild assert on the exact text of an error, so the prefix is DuckDB's spelling
104    /// including the parts that look like typos. `Not implemented Error` really is capitalised
105    /// that way upstream, and `INTERNAL Error` really is shouted.
106    #[must_use]
107    pub const fn duckdb_name(self) -> &'static str {
108        match self {
109            Self::Parser => "Parser Error",
110            Self::Syntax => "Syntax Error",
111            Self::Binder => "Binder Error",
112            Self::Catalog => "Catalog Error",
113            Self::Conversion => "Conversion Error",
114            Self::OutOfRange => "Out of Range Error",
115            Self::InvalidInput => "Invalid Input Error",
116            Self::OutOfMemory => "Out of Memory Error",
117            Self::Io => "IO Error",
118            Self::NotImplemented => "Not implemented Error",
119            Self::Constraint => "Constraint Error",
120            Self::Transaction => "TransactionContext Error",
121            Self::Settings => "Settings Error",
122            Self::Interrupt => "Interrupt Error",
123            Self::ParameterNotAllowed => "Parameter Not Allowed Error",
124            Self::MismatchType => "Mismatch Type Error",
125            Self::Internal => "INTERNAL Error",
126        }
127    }
128
129    /// Whether an error with this code says something about the query rather than about us.
130    ///
131    /// Used by the fuzzing harness in `spec/16-testing.md` section 16.4, which treats a rejected
132    /// query as a normal outcome and an internal error as a finding.
133    #[must_use]
134    pub const fn is_user_error(self) -> bool {
135        matches!(
136            self,
137            Self::Parser
138                | Self::Syntax
139                | Self::Binder
140                | Self::Catalog
141                | Self::Conversion
142                | Self::OutOfRange
143                | Self::InvalidInput
144                | Self::Constraint
145                | Self::Transaction
146                | Self::Settings
147                | Self::ParameterNotAllowed
148                | Self::MismatchType
149        )
150    }
151}
152
153impl fmt::Display for ErrorCode {
154    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
155        f.write_str(self.duckdb_name())
156    }
157}
158
159/// An error, carrying a code, a message and optionally where in the query it happened.
160///
161/// The payload is boxed so that `Error` is one pointer wide, which keeps `Result<T>` the same size
162/// as `T` for every `T` that has a niche. Errors are rare and results are returned from every
163/// function in the workspace, so the cost belongs on the rare path.
164#[derive(Debug, Clone, PartialEq, Eq)]
165pub struct Error(Box<Payload>);
166
167#[derive(Debug, Clone, PartialEq, Eq)]
168struct Payload {
169    code: ErrorCode,
170    message: String,
171    span: Option<Span>,
172    message_only: bool,
173}
174
175impl Error {
176    /// An error with a code and a message and no span.
177    pub fn new(code: ErrorCode, message: impl Into<String>) -> Self {
178        Self(Box::new(Payload { code, message: message.into(), span: None, message_only: false }))
179    }
180
181    /// The same error, with the part of the query it is about.
182    #[must_use]
183    pub fn with_span(mut self, span: Span) -> Self {
184        self.0.span = Some(span);
185        self
186    }
187
188    /// Attaches the range only when a more specific caller has not already attached one.
189    #[must_use]
190    pub fn with_fallback_span(mut self, span: Span) -> Self {
191        if self.0.span.is_none() && !span.is_empty() {
192            self.0.span = Some(span);
193        }
194        self
195    }
196
197    /// What kind of thing went wrong.
198    #[must_use]
199    pub fn code(&self) -> ErrorCode {
200        self.0.code
201    }
202
203    /// The message, without the code prefix that `Display` adds.
204    #[must_use]
205    pub fn message(&self) -> &str {
206        &self.0.message
207    }
208
209    /// Where in the query text this is about, if it is about a place.
210    #[must_use]
211    pub fn span(&self) -> Option<Span> {
212        self.0.span
213    }
214
215    /// Renders this error as the structured JSON form DuckDB returns for `errors_as_json`.
216    #[must_use]
217    pub fn into_json(mut self) -> Self {
218        let exception_type = self.0.code.json_name();
219        let subtype = self.0.code.json_subtype(&self.0.message);
220        let mut fields = vec![
221            ("exception_type", exception_type.to_string()),
222            ("exception_message", self.0.message.clone()),
223        ];
224        if let Some(span) = self.0.span {
225            fields.push(("location", format!("[{},{}]", span.start, span.len())));
226            fields.push(("position", span.start.to_string()));
227        }
228        if let Some(subtype) = subtype {
229            fields.push(("error_subtype", subtype.to_string()));
230        }
231        self.0.message = json_object(&fields);
232        self.0.message_only = true;
233        self
234    }
235
236    /// The text is not SQL.
237    pub fn parser(message: impl Into<String>) -> Self {
238        Self::new(ErrorCode::Parser, message)
239    }
240
241    /// The text is SQL and a value in it is not one the statement can take.
242    pub fn syntax(message: impl Into<String>) -> Self {
243        Self::new(ErrorCode::Syntax, message)
244    }
245
246    /// The text is SQL and it does not mean anything.
247    pub fn binder(message: impl Into<String>) -> Self {
248        Self::new(ErrorCode::Binder, message)
249    }
250
251    /// A named object is missing, or one that should be missing is not.
252    pub fn catalog(message: impl Into<String>) -> Self {
253        Self::new(ErrorCode::Catalog, message)
254    }
255
256    /// A value will not convert to the type it is being asked for.
257    pub fn conversion(message: impl Into<String>) -> Self {
258        Self::new(ErrorCode::Conversion, message)
259    }
260
261    /// A value is outside what its type can hold.
262    pub fn out_of_range(message: impl Into<String>) -> Self {
263        Self::new(ErrorCode::OutOfRange, message)
264    }
265
266    /// An argument is wrong in a way that is not a type error.
267    pub fn invalid_input(message: impl Into<String>) -> Self {
268        Self::new(ErrorCode::InvalidInput, message)
269    }
270
271    /// An allocation failed or a memory limit was reached.
272    pub fn out_of_memory(message: impl Into<String>) -> Self {
273        Self::new(ErrorCode::OutOfMemory, message)
274    }
275
276    /// The filesystem, the network or the object store said no.
277    pub fn io(message: impl Into<String>) -> Self {
278        Self::new(ErrorCode::Io, message)
279    }
280
281    /// It is in the plan and it is not built yet.
282    pub fn not_implemented(message: impl Into<String>) -> Self {
283        Self::new(ErrorCode::NotImplemented, message)
284    }
285
286    /// A constraint was violated.
287    pub fn constraint(message: impl Into<String>) -> Self {
288        Self::new(ErrorCode::Constraint, message)
289    }
290
291    /// A conflict, an abort, or a statement issued outside a transaction that needs one.
292    pub fn transaction(message: impl Into<String>) -> Self {
293        Self::new(ErrorCode::Transaction, message)
294    }
295
296    /// A setting cannot be applied in the current engine configuration.
297    pub fn settings(message: impl Into<String>) -> Self {
298        Self::new(ErrorCode::Settings, message)
299    }
300
301    /// A value the function refuses outright.
302    pub fn parameter_not_allowed(message: impl Into<String>) -> Self {
303        Self::new(ErrorCode::ParameterNotAllowed, message)
304    }
305
306    /// Two types that cannot meet.
307    pub fn mismatch_type(message: impl Into<String>) -> Self {
308        Self::new(ErrorCode::MismatchType, message)
309    }
310
311    /// The query was cancelled.
312    pub fn interrupt(message: impl Into<String>) -> Self {
313        Self::new(ErrorCode::Interrupt, message)
314    }
315
316    /// An invariant this code is responsible for does not hold.
317    ///
318    /// Reaching this is always a bug in the database and never a bug in the query, which is why it
319    /// reads differently from the others and why the fuzzer treats it as a finding.
320    pub fn internal(message: impl Into<String>) -> Self {
321        Self::new(ErrorCode::Internal, message)
322    }
323}
324
325impl fmt::Display for Error {
326    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
327        if self.0.message_only {
328            return f.write_str(&self.0.message);
329        }
330        write!(f, "{}: {}", self.0.code, self.0.message)
331    }
332}
333
334impl ErrorCode {
335    const fn json_name(self) -> &'static str {
336        match self {
337            Self::Parser => "Parser",
338            Self::Syntax => "Syntax",
339            Self::Binder => "Binder",
340            Self::Catalog => "Catalog",
341            Self::Conversion => "Conversion",
342            Self::OutOfRange => "Out of Range",
343            Self::InvalidInput => "Invalid Input",
344            Self::OutOfMemory => "Out of Memory",
345            Self::Io => "IO",
346            Self::NotImplemented => "Not implemented",
347            Self::Constraint => "Constraint",
348            Self::Transaction => "TransactionContext",
349            Self::Settings => "Settings",
350            Self::Interrupt => "Interrupt",
351            Self::ParameterNotAllowed => "Parameter Not Allowed",
352            Self::MismatchType => "Mismatch Type",
353            Self::Internal => "INTERNAL",
354        }
355    }
356
357    fn json_subtype(self, message: &str) -> Option<&'static str> {
358        match self {
359            Self::Parser => Some("SYNTAX_ERROR"),
360            Self::Binder if message.starts_with("Referenced column") => Some("COLUMN_NOT_FOUND"),
361            Self::Binder if message.contains("No function matches") => Some("NO_MATCHING_FUNCTION"),
362            Self::Catalog if message.contains("does not exist") => Some("MISSING_ENTRY"),
363            _ => None,
364        }
365    }
366}
367
368fn json_object(fields: &[(&str, String)]) -> String {
369    let mut out = String::from("{");
370    for (index, (name, value)) in fields.iter().enumerate() {
371        if index != 0 {
372            out.push(',');
373        }
374        out.push('"');
375        out.push_str(name);
376        out.push_str("\":\"");
377        for character in value.chars() {
378            match character {
379                '"' => out.push_str("\\\""),
380                '\\' => out.push_str("\\\\"),
381                '\n' => out.push_str("\\n"),
382                '\r' => out.push_str("\\r"),
383                '\t' => out.push_str("\\t"),
384                character if character <= '\u{1f}' => {
385                    use std::fmt::Write as _;
386                    let _ = write!(out, "\\u{:04x}", character as u32);
387                }
388                character => out.push(character),
389            }
390        }
391        out.push('"');
392    }
393    out.push('}');
394    out
395}
396
397impl std::error::Error for Error {}
398
399impl From<std::io::Error> for Error {
400    fn from(error: std::io::Error) -> Self {
401        Self::io(error.to_string())
402    }
403}
404
405#[cfg(test)]
406mod tests {
407    use super::{Error, ErrorCode, Span};
408
409    #[test]
410    fn an_error_prints_the_way_duckdb_prints_it() {
411        let error = Error::binder("Referenced column \"nope\" not found in FROM clause!");
412        assert_eq!(
413            error.to_string(),
414            "Binder Error: Referenced column \"nope\" not found in FROM clause!"
415        );
416    }
417
418    #[test]
419    fn a_json_error_is_structured_and_has_no_text_prefix() {
420        let error = Error::binder("Referenced column \"nope\" not found\nnext")
421            .with_span(Span::new(7, 11))
422            .into_json();
423        assert_eq!(
424            error.to_string(),
425            "{\"exception_type\":\"Binder\",\"exception_message\":\"Referenced column \\\"nope\\\" not found\\nnext\",\"location\":\"[7,4]\",\"position\":\"7\",\"error_subtype\":\"COLUMN_NOT_FOUND\"}"
426        );
427        assert_eq!(error.code(), ErrorCode::Binder);
428    }
429
430    #[test]
431    fn a_result_is_no_wider_than_the_value_in_it() {
432        // The reason the payload is boxed. If this ever fails, every function in the workspace
433        // got more expensive to return from and nobody noticed.
434        assert_eq!(size_of::<Error>(), size_of::<usize>());
435        assert_eq!(size_of::<Result<String, Error>>(), size_of::<String>());
436    }
437
438    #[test]
439    fn a_span_survives_being_attached() {
440        let error = Error::parser("syntax error at or near \"FROM\"").with_span(Span::new(7, 11));
441        assert_eq!(error.span(), Some(Span::new(7, 11)));
442        assert_eq!(error.span().map(Span::len), Some(4));
443        assert_eq!(error.code(), ErrorCode::Parser);
444    }
445
446    #[test]
447    fn a_fallback_span_keeps_the_more_specific_range() {
448        let specific = Error::binder("missing")
449            .with_span(Span::new(7, 14))
450            .with_fallback_span(Span::new(0, 20));
451        assert_eq!(specific.span(), Some(Span::new(7, 14)));
452        let fallback = Error::binder("missing").with_fallback_span(Span::new(0, 20));
453        assert_eq!(fallback.span(), Some(Span::new(0, 20)));
454        assert_eq!(Error::binder("missing").with_fallback_span(Span::new(0, 0)).span(), None);
455    }
456
457    #[test]
458    fn the_fuzzer_can_tell_our_bugs_from_the_query_s_bugs() {
459        assert!(ErrorCode::Binder.is_user_error());
460        assert!(ErrorCode::Conversion.is_user_error());
461        assert!(!ErrorCode::Internal.is_user_error());
462        assert!(!ErrorCode::OutOfMemory.is_user_error());
463        // Not implemented is ours rather than the query's, because the query was legitimate and we
464        // are the reason it did not run.
465        assert!(!ErrorCode::NotImplemented.is_user_error());
466    }
467}
rudb_common/error.rs

rudb_common/
error.rs
