Skip to main content

lemma/parsing/
parser.rs

1use crate::error::Error;
2use crate::limits::ResourceLimits;
3use crate::parsing::ast::{try_parse_type_constraint_command, *};
4use crate::parsing::lexer::{
5    can_be_label, can_be_reference_segment, conversion_target_from_token, is_boolean_keyword,
6    is_calendar_unit_token, is_duration_unit, is_math_function, is_spec_body_keyword,
7    is_structural_keyword, is_type_keyword, token_kind_to_boolean_value,
8    token_kind_to_calendar_unit, token_kind_to_duration_unit, token_kind_to_primitive, Lexer,
9    Token, TokenKind,
10};
11use crate::parsing::source::Source;
12use rust_decimal::Decimal;
13use std::str::FromStr;
14use std::sync::Arc;
15
16type TypeArrowChain = (ParentType, Option<SpecRef>, Option<Vec<Constraint>>);
17
18pub struct ParseResult {
19    pub specs: Vec<LemmaSpec>,
20    pub expression_count: usize,
21}
22
23pub fn parse(
24    content: &str,
25    attribute: &str,
26    limits: &ResourceLimits,
27) -> Result<ParseResult, Error> {
28    if content.len() > limits.max_file_size_bytes {
29        return Err(Error::resource_limit_exceeded(
30            "max_file_size_bytes",
31            format!(
32                "{} bytes ({} MB)",
33                limits.max_file_size_bytes,
34                limits.max_file_size_bytes / (1024 * 1024)
35            ),
36            format!(
37                "{} bytes ({:.2} MB)",
38                content.len(),
39                content.len() as f64 / (1024.0 * 1024.0)
40            ),
41            "Reduce file size or split into multiple specs",
42            None,
43            None,
44            None,
45        ));
46    }
47
48    let mut parser = Parser::new(content, attribute, limits);
49    let specs = parser.parse_file()?;
50    Ok(ParseResult {
51        specs,
52        expression_count: parser.expression_count,
53    })
54}
55
56struct Parser {
57    lexer: Lexer,
58    depth_tracker: DepthTracker,
59    expression_count: usize,
60    max_expression_count: usize,
61}
62
63impl Parser {
64    fn new(content: &str, attribute: &str, limits: &ResourceLimits) -> Self {
65        Parser {
66            lexer: Lexer::new(content, attribute),
67            depth_tracker: DepthTracker::with_max_depth(limits.max_expression_depth),
68            expression_count: 0,
69            max_expression_count: limits.max_expression_count,
70        }
71    }
72
73    fn attribute(&self) -> String {
74        self.lexer.attribute().to_string()
75    }
76
77    fn peek(&mut self) -> Result<&Token, Error> {
78        self.lexer.peek()
79    }
80
81    fn next(&mut self) -> Result<Token, Error> {
82        self.lexer.next_token()
83    }
84
85    fn at(&mut self, kind: &TokenKind) -> Result<bool, Error> {
86        Ok(&self.peek()?.kind == kind)
87    }
88
89    fn at_any(&mut self, kinds: &[TokenKind]) -> Result<bool, Error> {
90        let current = &self.peek()?.kind;
91        Ok(kinds.contains(current))
92    }
93
94    fn expect(&mut self, kind: &TokenKind) -> Result<Token, Error> {
95        let token = self.next()?;
96        if &token.kind == kind {
97            Ok(token)
98        } else {
99            Err(self.error_at_token(&token, format!("Expected {}, found {}", kind, token.kind)))
100        }
101    }
102
103    fn error_at_token(&self, token: &Token, message: impl Into<String>) -> Error {
104        Error::parsing(
105            message,
106            Source::new(self.lexer.attribute(), token.span.clone()),
107            None::<String>,
108        )
109    }
110
111    fn error_at_token_with_suggestion(
112        &self,
113        token: &Token,
114        message: impl Into<String>,
115        suggestion: impl Into<String>,
116    ) -> Error {
117        Error::parsing(
118            message,
119            Source::new(self.lexer.attribute(), token.span.clone()),
120            Some(suggestion),
121        )
122    }
123
124    /// Parse `~ HASH` where HASH is 8 alphanumeric chars. Optional — returns
125    /// `Ok(None)` when the next token is not `~`. Uses raw scanning after the
126    /// tilde to bypass tokenization (avoids scientific-notation mis-lexing of
127    /// hashes like `7e20848b`).
128    fn try_parse_hash_pin(&mut self) -> Result<Option<String>, Error> {
129        if !self.at(&TokenKind::Tilde)? {
130            return Ok(None);
131        }
132        let tilde_span = self.next()?.span;
133        let hash = self.lexer.scan_raw_alphanumeric()?;
134        if hash.len() != 8 {
135            return Err(Error::parsing(
136                format!(
137                    "Expected an 8-character alphanumeric plan hash after '~', found '{}'",
138                    hash
139                ),
140                self.make_source(tilde_span),
141                None::<String>,
142            ));
143        }
144        Ok(Some(hash))
145    }
146
147    fn make_source(&self, span: Span) -> Source {
148        Source::new(self.lexer.attribute(), span)
149    }
150
151    fn span_from(&self, start: &Span) -> Span {
152        // Create a span from start to the current lexer position.
153        // We peek to get the current position.
154        Span {
155            start: start.start,
156            end: start.end.max(start.start),
157            line: start.line,
158            col: start.col,
159        }
160    }
161
162    fn span_covering(&self, start: &Span, end: &Span) -> Span {
163        Span {
164            start: start.start,
165            end: end.end,
166            line: start.line,
167            col: start.col,
168        }
169    }
170
171    // ========================================================================
172    // Top-level: file and spec
173    // ========================================================================
174
175    fn parse_file(&mut self) -> Result<Vec<LemmaSpec>, Error> {
176        let mut specs = Vec::new();
177        loop {
178            if self.at(&TokenKind::Eof)? {
179                break;
180            }
181            if self.at(&TokenKind::Spec)? {
182                specs.push(self.parse_spec()?);
183            } else {
184                let token = self.next()?;
185                return Err(self.error_at_token_with_suggestion(
186                    &token,
187                    format!(
188                        "Expected a spec declaration (e.g. 'spec my_spec'), found {}",
189                        token.kind
190                    ),
191                    "A Lemma file must start with 'spec <name>'",
192                ));
193            }
194        }
195        Ok(specs)
196    }
197
198    fn parse_spec(&mut self) -> Result<LemmaSpec, Error> {
199        let spec_token = self.expect(&TokenKind::Spec)?;
200        let start_line = spec_token.span.line;
201
202        let (name, _name_span) = self.parse_spec_name()?;
203
204        let effective_from = self.try_parse_effective_from()?;
205
206        let commentary = self.try_parse_commentary()?;
207
208        let attribute = self.attribute();
209        let mut spec = LemmaSpec::new(name.clone())
210            .with_attribute(attribute)
211            .with_start_line(start_line);
212        spec.effective_from = effective_from;
213
214        if let Some(commentary_text) = commentary {
215            spec = spec.set_commentary(commentary_text);
216        }
217
218        // First pass: collect type definitions
219        // We need to peek and handle type definitions first, but since we consume tokens
220        // linearly, we'll collect all items in one pass.
221        let mut facts = Vec::new();
222        let mut rules = Vec::new();
223        let mut types = Vec::new();
224        let mut meta_fields = Vec::new();
225
226        loop {
227            let peek_kind = self.peek()?.kind.clone();
228            match peek_kind {
229                TokenKind::Fact => {
230                    let fact = self.parse_fact()?;
231                    facts.push(fact);
232                }
233                TokenKind::Rule => {
234                    let rule = self.parse_rule()?;
235                    rules.push(rule);
236                }
237                TokenKind::Type => {
238                    let type_def = self.parse_type_def()?;
239                    types.push(type_def);
240                }
241                TokenKind::Meta => {
242                    let meta = self.parse_meta()?;
243                    meta_fields.push(meta);
244                }
245                TokenKind::Spec | TokenKind::Eof => break,
246                _ => {
247                    let token = self.next()?;
248                    return Err(self.error_at_token_with_suggestion(
249                        &token,
250                        format!(
251                            "Expected 'fact', 'rule', 'type', 'meta', or a new 'spec', found '{}'",
252                            token.text
253                        ),
254                        "Check the spelling or add the appropriate keyword",
255                    ));
256                }
257            }
258        }
259
260        for type_def in types {
261            spec = spec.add_type(type_def);
262        }
263        for fact in facts {
264            spec = spec.add_fact(fact);
265        }
266        for rule in rules {
267            spec = spec.add_rule(rule);
268        }
269        for meta in meta_fields {
270            spec = spec.add_meta_field(meta);
271        }
272
273        Ok(spec)
274    }
275
276    /// Parse a spec name: optional @ prefix, then identifier segments separated by /
277    /// Allows: "myspec", "contracts/employment/jack", "@user/workspace/spec"
278    fn parse_spec_name(&mut self) -> Result<(String, Span), Error> {
279        let mut name = String::new();
280        let start_span;
281
282        if self.at(&TokenKind::At)? {
283            let at_tok = self.next()?;
284            start_span = at_tok.span.clone();
285            name.push('@');
286        } else {
287            start_span = self.peek()?.span.clone();
288        }
289
290        // First segment must be an identifier or a keyword that can serve as name
291        let first = self.next()?;
292        if !first.kind.is_identifier_like() {
293            return Err(self.error_at_token(
294                &first,
295                format!("Expected a spec name, found {}", first.kind),
296            ));
297        }
298        name.push_str(&first.text);
299        let mut end_span = first.span.clone();
300
301        // Continue consuming / identifier segments
302        while self.at(&TokenKind::Slash)? {
303            self.next()?; // consume /
304            name.push('/');
305            let seg = self.next()?;
306            if !seg.kind.is_identifier_like() {
307                return Err(self.error_at_token(
308                    &seg,
309                    format!(
310                        "Expected identifier after '/' in spec name, found {}",
311                        seg.kind
312                    ),
313                ));
314            }
315            name.push_str(&seg.text);
316            end_span = seg.span.clone();
317        }
318
319        // Check for hyphen-containing spec names like "my-spec"
320        while self.at(&TokenKind::Minus)? {
321            // Only consume if the next token after minus is an identifier
322            // (hyphenated names like "my-spec")
323            let minus_span = self.peek()?.span.clone();
324            self.next()?; // consume -
325            if let Ok(peeked) = self.peek() {
326                if peeked.kind.is_identifier_like() {
327                    let seg = self.next()?;
328                    name.push('-');
329                    name.push_str(&seg.text);
330                    end_span = seg.span.clone();
331                    // Could be followed by /
332                    while self.at(&TokenKind::Slash)? {
333                        self.next()?; // consume /
334                        name.push('/');
335                        let seg2 = self.next()?;
336                        if !seg2.kind.is_identifier_like() {
337                            return Err(self.error_at_token(
338                                &seg2,
339                                format!(
340                                    "Expected identifier after '/' in spec name, found {}",
341                                    seg2.kind
342                                ),
343                            ));
344                        }
345                        name.push_str(&seg2.text);
346                        end_span = seg2.span.clone();
347                    }
348                } else {
349                    // The minus wasn't part of the name; this is an error
350                    let span = self.span_covering(&start_span, &minus_span);
351                    return Err(Error::parsing(
352                        "Trailing '-' after spec name",
353                        self.make_source(span),
354                        None::<String>,
355                    ));
356                }
357            }
358        }
359
360        let full_span = self.span_covering(&start_span, &end_span);
361        Ok((name, full_span))
362    }
363
364    fn try_parse_effective_from(&mut self) -> Result<Option<DateTimeValue>, Error> {
365        // effective_from is a date/time token right after the spec name.
366        // It's tricky because it looks like a number (e.g. 2026-03-04).
367        // In the old grammar it was a special atomic rule.
368        // We'll check if the next token is a NumberLit that looks like a year.
369        if !self.at(&TokenKind::NumberLit)? {
370            return Ok(None);
371        }
372
373        let peeked = self.peek()?;
374        let peeked_text = peeked.text.clone();
375        let peeked_span = peeked.span.clone();
376
377        // Check if it could be a date: 4-digit number followed by -
378        if peeked_text.len() == 4 && peeked_text.chars().all(|c| c.is_ascii_digit()) {
379            // Collect the full datetime string by consuming tokens
380            let mut dt_str = String::new();
381            let num_tok = self.next()?; // consume the year number
382            dt_str.push_str(&num_tok.text);
383
384            // Try to consume -MM-DD and optional T... parts
385            while self.at(&TokenKind::Minus)? {
386                self.next()?; // consume -
387                dt_str.push('-');
388                let part = self.next()?;
389                dt_str.push_str(&part.text);
390            }
391
392            // Check for T (time part)
393            if self.at(&TokenKind::Identifier)? {
394                let peeked = self.peek()?;
395                if peeked.text.starts_with('T') || peeked.text.starts_with('t') {
396                    let time_part = self.next()?;
397                    dt_str.push_str(&time_part.text);
398                    // Consume any : separated parts
399                    while self.at(&TokenKind::Colon)? {
400                        self.next()?;
401                        dt_str.push(':');
402                        let part = self.next()?;
403                        dt_str.push_str(&part.text);
404                    }
405                    // Check for timezone (+ or Z)
406                    if self.at(&TokenKind::Plus)? {
407                        self.next()?;
408                        dt_str.push('+');
409                        let tz_part = self.next()?;
410                        dt_str.push_str(&tz_part.text);
411                        if self.at(&TokenKind::Colon)? {
412                            self.next()?;
413                            dt_str.push(':');
414                            let tz_min = self.next()?;
415                            dt_str.push_str(&tz_min.text);
416                        }
417                    }
418                }
419            }
420
421            // Try to parse as datetime
422            if let Ok(dtv) = dt_str.parse::<DateTimeValue>() {
423                return Ok(Some(dtv));
424            }
425
426            return Err(Error::parsing(
427                format!("Invalid date/time in spec declaration: '{}'", dt_str),
428                self.make_source(peeked_span),
429                None::<String>,
430            ));
431        }
432
433        Ok(None)
434    }
435
436    fn try_parse_commentary(&mut self) -> Result<Option<String>, Error> {
437        if !self.at(&TokenKind::Commentary)? {
438            return Ok(None);
439        }
440        let token = self.next()?;
441        let trimmed = token.text.trim().to_string();
442        if trimmed.is_empty() {
443            Ok(None)
444        } else {
445            Ok(Some(trimmed))
446        }
447    }
448
449    // ========================================================================
450    // Fact parsing
451    // ========================================================================
452
453    fn parse_fact(&mut self) -> Result<LemmaFact, Error> {
454        let fact_token = self.expect(&TokenKind::Fact)?;
455        let start_span = fact_token.span.clone();
456
457        // Parse fact reference (single segment = definition, multi-segment = binding)
458        let reference = self.parse_reference()?;
459
460        self.expect(&TokenKind::Colon)?;
461
462        let value = self.parse_fact_value()?;
463
464        let end_span = self.peek()?.span.clone();
465        let span = self.span_covering(&start_span, &end_span);
466        let source = self.make_source(span);
467
468        Ok(LemmaFact::new(reference, value, source))
469    }
470
471    fn parse_reference(&mut self) -> Result<Reference, Error> {
472        let mut segments = Vec::new();
473
474        let first = self.next()?;
475        // Structural keywords (spec, fact, rule, unless, ...) cannot be names.
476        // Type keywords (duration, number, date, ...) CAN be names per the grammar.
477        if is_structural_keyword(&first.kind) {
478            return Err(self.error_at_token_with_suggestion(
479                &first,
480                format!(
481                    "'{}' is a reserved keyword and cannot be used as a name",
482                    first.text
483                ),
484                "Choose a different name that is not a reserved keyword",
485            ));
486        }
487
488        if !can_be_reference_segment(&first.kind) {
489            return Err(self.error_at_token(
490                &first,
491                format!("Expected an identifier, found {}", first.kind),
492            ));
493        }
494
495        segments.push(first.text.clone());
496
497        // Consume . separated segments
498        while self.at(&TokenKind::Dot)? {
499            self.next()?; // consume .
500            let seg = self.next()?;
501            if !can_be_reference_segment(&seg.kind) {
502                return Err(self.error_at_token(
503                    &seg,
504                    format!("Expected an identifier after '.', found {}", seg.kind),
505                ));
506            }
507            segments.push(seg.text.clone());
508        }
509
510        Ok(Reference::from_path(segments))
511    }
512
513    fn parse_fact_value(&mut self) -> Result<FactValue, Error> {
514        // Check for type declaration: [type_name] or [type_arrow_chain]
515        if self.at(&TokenKind::LBracket)? {
516            return self.parse_type_declaration_or_inline();
517        }
518
519        // Check for spec reference: spec <name>
520        if self.at(&TokenKind::Spec)? {
521            return self.parse_fact_spec_reference();
522        }
523
524        // Otherwise, it's a literal value
525        let value = self.parse_literal_value()?;
526        Ok(FactValue::Literal(value))
527    }
528
529    fn parse_type_declaration_or_inline(&mut self) -> Result<FactValue, Error> {
530        self.expect(&TokenKind::LBracket)?;
531
532        // Parse the type name (could be a standard type or custom type)
533        let (base, from_spec, constraints) = self.parse_type_arrow_chain()?;
534
535        self.expect(&TokenKind::RBracket)?;
536
537        Ok(FactValue::TypeDeclaration {
538            base,
539            constraints,
540            from: from_spec,
541        })
542    }
543
544    fn parse_fact_spec_reference(&mut self) -> Result<FactValue, Error> {
545        self.expect(&TokenKind::Spec)?;
546
547        let (name, _name_span) = self.parse_spec_name()?;
548        let from_registry = name.starts_with('@');
549
550        let hash_pin = self.try_parse_hash_pin()?;
551
552        let mut effective = None;
553        // Check for effective datetime after spec reference
554        if self.at(&TokenKind::NumberLit)? {
555            let peeked = self.peek()?;
556            if peeked.text.len() == 4 && peeked.text.chars().all(|c| c.is_ascii_digit()) {
557                // Could be a datetime effective
558                effective = self.try_parse_effective_from()?;
559            }
560        }
561
562        Ok(FactValue::SpecReference(SpecRef {
563            name,
564            from_registry,
565            hash_pin,
566            effective,
567        }))
568    }
569
570    // ========================================================================
571    // Rule parsing
572    // ========================================================================
573
574    fn parse_rule(&mut self) -> Result<LemmaRule, Error> {
575        let rule_token = self.expect(&TokenKind::Rule)?;
576        let start_span = rule_token.span.clone();
577
578        let name_tok = self.next()?;
579        if is_structural_keyword(&name_tok.kind) {
580            return Err(self.error_at_token_with_suggestion(
581                &name_tok,
582                format!(
583                    "'{}' is a reserved keyword and cannot be used as a rule name",
584                    name_tok.text
585                ),
586                "Choose a different name that is not a reserved keyword",
587            ));
588        }
589        if !can_be_label(&name_tok.kind) && !is_type_keyword(&name_tok.kind) {
590            return Err(self.error_at_token(
591                &name_tok,
592                format!("Expected a rule name, found {}", name_tok.kind),
593            ));
594        }
595        let rule_name = name_tok.text.clone();
596
597        self.expect(&TokenKind::Colon)?;
598
599        // Parse the base expression or veto
600        let expression = if self.at(&TokenKind::Veto)? {
601            self.parse_veto_expression()?
602        } else {
603            self.parse_expression()?
604        };
605
606        // Parse unless clauses
607        let mut unless_clauses = Vec::new();
608        while self.at(&TokenKind::Unless)? {
609            unless_clauses.push(self.parse_unless_clause()?);
610        }
611
612        let end_span = if let Some(last_unless) = unless_clauses.last() {
613            last_unless.source_location.span.clone()
614        } else if let Some(ref loc) = expression.source_location {
615            loc.span.clone()
616        } else {
617            start_span.clone()
618        };
619
620        let span = self.span_covering(&start_span, &end_span);
621        Ok(LemmaRule {
622            name: rule_name,
623            expression,
624            unless_clauses,
625            source_location: self.make_source(span),
626        })
627    }
628
629    fn parse_veto_expression(&mut self) -> Result<Expression, Error> {
630        let veto_tok = self.expect(&TokenKind::Veto)?;
631        let start_span = veto_tok.span.clone();
632
633        let message = if self.at(&TokenKind::StringLit)? {
634            let str_tok = self.next()?;
635            let content = unquote_string(&str_tok.text);
636            Some(content)
637        } else {
638            None
639        };
640
641        let span = self.span_from(&start_span);
642        self.new_expression(
643            ExpressionKind::Veto(VetoExpression { message }),
644            self.make_source(span),
645        )
646    }
647
648    fn parse_unless_clause(&mut self) -> Result<UnlessClause, Error> {
649        let unless_tok = self.expect(&TokenKind::Unless)?;
650        let start_span = unless_tok.span.clone();
651
652        let condition = self.parse_expression()?;
653
654        self.expect(&TokenKind::Then)?;
655
656        let result = if self.at(&TokenKind::Veto)? {
657            self.parse_veto_expression()?
658        } else {
659            self.parse_expression()?
660        };
661
662        let end_span = result
663            .source_location
664            .as_ref()
665            .map(|s| s.span.clone())
666            .unwrap_or_else(|| start_span.clone());
667        let span = self.span_covering(&start_span, &end_span);
668
669        Ok(UnlessClause {
670            condition,
671            result,
672            source_location: self.make_source(span),
673        })
674    }
675
676    // ========================================================================
677    // Type definitions
678    // ========================================================================
679
680    fn parse_type_def(&mut self) -> Result<TypeDef, Error> {
681        let type_tok = self.expect(&TokenKind::Type)?;
682        let start_span = type_tok.span.clone();
683
684        // Parse type name
685        let name_tok = self.next()?;
686        let type_name = name_tok.text.clone();
687
688        // Check if this is a type import (type X from Y) or a type definition (type X: Y)
689        if self.at(&TokenKind::From)? {
690            return self.parse_type_import(type_name, start_span);
691        }
692
693        // Regular type definition: type X: Y -> ...
694        if self.at(&TokenKind::Colon)? {
695            self.next()?; // consume :
696        } else {
697            // Could also be an import without us seeing 'from' yet if there are two type names
698            // e.g. "type money from other_spec"
699            let peek = self.peek()?.clone();
700            return Err(self.error_at_token(
701                &peek,
702                format!(
703                    "Expected ':' or 'from' after type name '{}', found {}",
704                    type_name, peek.kind
705                ),
706            ));
707        }
708
709        let (parent, _from, constraints) = self.parse_type_arrow_chain()?;
710
711        let end_span = self.peek()?.span.clone();
712        let span = self.span_covering(&start_span, &end_span);
713        Ok(TypeDef::Regular {
714            source_location: self.make_source(span),
715            name: type_name,
716            parent,
717            constraints,
718        })
719    }
720
721    fn parse_type_import(&mut self, type_name: String, start_span: Span) -> Result<TypeDef, Error> {
722        self.expect(&TokenKind::From)?;
723
724        let (from_name, _from_span) = self.parse_spec_name()?;
725        let from_registry = from_name.starts_with('@');
726        let hash_pin = self.try_parse_hash_pin()?;
727
728        let from = SpecRef {
729            name: from_name,
730            from_registry,
731            hash_pin,
732            effective: None,
733        };
734
735        // Check for arrow chain constraints after import
736        let constraints = if self.at(&TokenKind::Arrow)? {
737            let (_, _, constraints) = self.parse_remaining_arrow_chain()?;
738            constraints
739        } else {
740            None
741        };
742
743        let end_span = self.peek()?.span.clone();
744        let span = self.span_covering(&start_span, &end_span);
745
746        let source_type = type_name.clone();
747
748        Ok(TypeDef::Import {
749            source_location: self.make_source(span),
750            name: type_name,
751            source_type,
752            from,
753            constraints,
754        })
755    }
756
757    /// Parse a type arrow chain: type_name (-> command)* or type_name from spec (-> command)*
758    fn parse_type_arrow_chain(&mut self) -> Result<TypeArrowChain, Error> {
759        let name_tok = self.next()?;
760        let base = if let Some(kind) = token_kind_to_primitive(&name_tok.kind) {
761            ParentType::Primitive { primitive: kind }
762        } else if can_be_label(&name_tok.kind) {
763            ParentType::Custom {
764                name: name_tok.text.clone(),
765            }
766        } else {
767            return Err(self.error_at_token(
768                &name_tok,
769                format!("Expected a type name, found {}", name_tok.kind),
770            ));
771        };
772
773        // Check for 'from' (inline type import)
774        let from_spec = if self.at(&TokenKind::From)? {
775            self.next()?; // consume from
776            let (from_name, _) = self.parse_spec_name()?;
777            let from_registry = from_name.starts_with('@');
778            let hash_pin = self.try_parse_hash_pin()?;
779            Some(SpecRef {
780                name: from_name,
781                from_registry,
782                hash_pin,
783                effective: None,
784            })
785        } else {
786            None
787        };
788
789        // Parse arrow chain constraints
790        let mut commands = Vec::new();
791        while self.at(&TokenKind::Arrow)? {
792            self.next()?; // consume ->
793            let (cmd, cmd_args) = self.parse_command()?;
794            commands.push((cmd, cmd_args));
795        }
796
797        let constraints = if commands.is_empty() {
798            None
799        } else {
800            Some(commands)
801        };
802
803        Ok((base, from_spec, constraints))
804    }
805
806    fn parse_remaining_arrow_chain(&mut self) -> Result<TypeArrowChain, Error> {
807        let mut commands = Vec::new();
808        while self.at(&TokenKind::Arrow)? {
809            self.next()?; // consume ->
810            let (cmd, cmd_args) = self.parse_command()?;
811            commands.push((cmd, cmd_args));
812        }
813        let constraints = if commands.is_empty() {
814            None
815        } else {
816            Some(commands)
817        };
818        Ok((
819            ParentType::Custom {
820                name: String::new(),
821            },
822            None,
823            constraints,
824        ))
825    }
826
827    fn parse_command(&mut self) -> Result<(TypeConstraintCommand, Vec<CommandArg>), Error> {
828        let name_tok = self.next()?;
829        if !can_be_label(&name_tok.kind) && !is_type_keyword(&name_tok.kind) {
830            return Err(self.error_at_token(
831                &name_tok,
832                format!("Expected a command name, found {}", name_tok.kind),
833            ));
834        }
835        let cmd = try_parse_type_constraint_command(&name_tok.text).ok_or_else(|| {
836            self.error_at_token(
837                &name_tok,
838                format!(
839                    "Unknown constraint command '{}'. Valid commands: help, default, unit, minimum, maximum, decimals, precision, option, options, length",
840                    name_tok.text
841                ),
842            )
843        })?;
844
845        let mut args = Vec::new();
846        loop {
847            // Command args: number, boolean, text, or label
848            // Stop at: ->, ], newlines (next keyword), EOF
849            if self.at(&TokenKind::Arrow)?
850                || self.at(&TokenKind::RBracket)?
851                || self.at(&TokenKind::Eof)?
852                || is_spec_body_keyword(&self.peek()?.kind)
853                || self.at(&TokenKind::Spec)?
854            {
855                break;
856            }
857
858            let peek_kind = self.peek()?.kind.clone();
859            match peek_kind {
860                TokenKind::NumberLit => {
861                    let tok = self.next()?;
862                    args.push(CommandArg::Number(tok.text));
863                }
864                TokenKind::Minus | TokenKind::Plus => {
865                    let second = self.lexer.peek_second()?.kind.clone();
866                    if second == TokenKind::NumberLit {
867                        let sign = self.next()?;
868                        let num = self.next()?;
869                        let text = format!("{}{}", sign.text, num.text);
870                        args.push(CommandArg::Number(text));
871                    } else {
872                        break;
873                    }
874                }
875                TokenKind::StringLit => {
876                    let tok = self.next()?;
877                    let content = unquote_string(&tok.text);
878                    args.push(CommandArg::Text(content));
879                }
880                ref k if is_boolean_keyword(k) => {
881                    let tok = self.next()?;
882                    args.push(CommandArg::Boolean(token_kind_to_boolean_value(&tok.kind)));
883                }
884                ref k if can_be_label(k) || is_type_keyword(k) => {
885                    let tok = self.next()?;
886                    args.push(CommandArg::Label(tok.text));
887                }
888                _ => break,
889            }
890        }
891
892        Ok((cmd, args))
893    }
894
895    // ========================================================================
896    // Meta parsing
897    // ========================================================================
898
899    fn parse_meta(&mut self) -> Result<MetaField, Error> {
900        let meta_tok = self.expect(&TokenKind::Meta)?;
901        let start_span = meta_tok.span.clone();
902
903        let key_tok = self.next()?;
904        let key = key_tok.text.clone();
905
906        self.expect(&TokenKind::Colon)?;
907
908        let value = self.parse_meta_value()?;
909
910        let end_span = self.peek()?.span.clone();
911        let span = self.span_covering(&start_span, &end_span);
912
913        Ok(MetaField {
914            key,
915            value,
916            source_location: self.make_source(span),
917        })
918    }
919
920    fn parse_meta_value(&mut self) -> Result<MetaValue, Error> {
921        // Try literal first (string, number, boolean, date)
922        let peeked = self.peek()?;
923        match &peeked.kind {
924            TokenKind::StringLit => {
925                let value = self.parse_literal_value()?;
926                return Ok(MetaValue::Literal(value));
927            }
928            TokenKind::NumberLit => {
929                let value = self.parse_literal_value()?;
930                return Ok(MetaValue::Literal(value));
931            }
932            k if is_boolean_keyword(k) => {
933                let value = self.parse_literal_value()?;
934                return Ok(MetaValue::Literal(value));
935            }
936            _ => {}
937        }
938
939        // Otherwise, consume as unquoted meta identifier
940        // meta_identifier: (ASCII_ALPHANUMERIC | "_" | "-" | "." | "/")+
941        let mut ident = String::new();
942        loop {
943            let peeked = self.peek()?;
944            match &peeked.kind {
945                k if k.is_identifier_like() => {
946                    let tok = self.next()?;
947                    ident.push_str(&tok.text);
948                }
949                TokenKind::Dot => {
950                    self.next()?;
951                    ident.push('.');
952                }
953                TokenKind::Slash => {
954                    self.next()?;
955                    ident.push('/');
956                }
957                TokenKind::Minus => {
958                    self.next()?;
959                    ident.push('-');
960                }
961                TokenKind::NumberLit => {
962                    let tok = self.next()?;
963                    ident.push_str(&tok.text);
964                }
965                _ => break,
966            }
967        }
968
969        if ident.is_empty() {
970            let tok = self.peek()?.clone();
971            return Err(self.error_at_token(&tok, "Expected a meta value"));
972        }
973
974        Ok(MetaValue::Unquoted(ident))
975    }
976
977    // ========================================================================
978    // Literal value parsing
979    // ========================================================================
980
981    fn parse_literal_value(&mut self) -> Result<Value, Error> {
982        let peeked = self.peek()?;
983        match &peeked.kind {
984            TokenKind::StringLit => {
985                let tok = self.next()?;
986                let content = unquote_string(&tok.text);
987                Ok(Value::Text(content))
988            }
989            k if is_boolean_keyword(k) => {
990                let tok = self.next()?;
991                Ok(Value::Boolean(token_kind_to_boolean_value(&tok.kind)))
992            }
993            TokenKind::NumberLit => self.parse_number_literal(),
994            TokenKind::Minus | TokenKind::Plus => self.parse_signed_number_literal(),
995            _ => {
996                let tok = self.next()?;
997                Err(self.error_at_token(
998                    &tok,
999                    format!(
1000                        "Expected a value (number, text, boolean, date, etc.), found '{}'",
1001                        tok.text
1002                    ),
1003                ))
1004            }
1005        }
1006    }
1007
1008    fn parse_signed_number_literal(&mut self) -> Result<Value, Error> {
1009        let sign_tok = self.next()?;
1010        let sign_span = sign_tok.span.clone();
1011        let is_negative = sign_tok.kind == TokenKind::Minus;
1012
1013        if !self.at(&TokenKind::NumberLit)? {
1014            let tok = self.peek()?.clone();
1015            return Err(self.error_at_token(
1016                &tok,
1017                format!(
1018                    "Expected a number after '{}', found '{}'",
1019                    sign_tok.text, tok.text
1020                ),
1021            ));
1022        }
1023
1024        let value = self.parse_number_literal()?;
1025        if !is_negative {
1026            return Ok(value);
1027        }
1028        match value {
1029            Value::Number(d) => Ok(Value::Number(-d)),
1030            Value::Scale(d, unit) => Ok(Value::Scale(-d, unit)),
1031            Value::Duration(d, unit) => Ok(Value::Duration(-d, unit)),
1032            Value::Ratio(d, label) => Ok(Value::Ratio(-d, label)),
1033            other => Err(Error::parsing(
1034                format!("Cannot negate this value: {}", other),
1035                self.make_source(sign_span),
1036                None::<String>,
1037            )),
1038        }
1039    }
1040
1041    fn parse_number_literal(&mut self) -> Result<Value, Error> {
1042        let num_tok = self.next()?;
1043        let num_text = &num_tok.text;
1044        let num_span = num_tok.span.clone();
1045
1046        // Check if followed by - which could make it a date (YYYY-MM-DD)
1047        if num_text.len() == 4
1048            && num_text.chars().all(|c| c.is_ascii_digit())
1049            && self.at(&TokenKind::Minus)?
1050        {
1051            return self.parse_date_literal(num_text.clone(), num_span);
1052        }
1053
1054        // Check what follows the number
1055        let peeked = self.peek()?;
1056
1057        // Number followed by : could be a time literal (HH:MM:SS)
1058        if num_text.len() == 2
1059            && num_text.chars().all(|c| c.is_ascii_digit())
1060            && peeked.kind == TokenKind::Colon
1061        {
1062            // Only if we're in a fact value context... this is ambiguous.
1063            // Time literals look like: 14:30:00 or 14:30
1064            // But we might also have "rule x: expr" where : is assignment.
1065            // The grammar handles this at the grammar level. For us,
1066            // we need to check if the context is right.
1067            // Let's try to parse as time if the following pattern matches.
1068            return self.try_parse_time_literal(num_text.clone(), num_span);
1069        }
1070
1071        // Check for %% (permille) - must be before %
1072        if peeked.kind == TokenKind::PercentPercent {
1073            let pp_tok = self.next()?;
1074            // Check it's not followed by a digit
1075            if let Ok(next_peek) = self.peek() {
1076                if next_peek.kind == TokenKind::NumberLit {
1077                    return Err(self.error_at_token(
1078                        &pp_tok,
1079                        "Permille literal cannot be followed by a digit",
1080                    ));
1081                }
1082            }
1083            let decimal = parse_decimal_string(num_text, &num_span, self)?;
1084            let ratio_value = decimal / Decimal::from(1000);
1085            return Ok(Value::Ratio(ratio_value, Some("permille".to_string())));
1086        }
1087
1088        // Check for % (percent)
1089        if peeked.kind == TokenKind::Percent {
1090            let pct_tok = self.next()?;
1091            // Check it's not followed by a digit or another %
1092            if let Ok(next_peek) = self.peek() {
1093                if next_peek.kind == TokenKind::NumberLit || next_peek.kind == TokenKind::Percent {
1094                    return Err(self.error_at_token(
1095                        &pct_tok,
1096                        "Percent literal cannot be followed by a digit",
1097                    ));
1098                }
1099            }
1100            let decimal = parse_decimal_string(num_text, &num_span, self)?;
1101            let ratio_value = decimal / Decimal::from(100);
1102            return Ok(Value::Ratio(ratio_value, Some("percent".to_string())));
1103        }
1104
1105        // Check for "percent" keyword
1106        if peeked.kind == TokenKind::PercentKw {
1107            self.next()?; // consume "percent"
1108            let decimal = parse_decimal_string(num_text, &num_span, self)?;
1109            let ratio_value = decimal / Decimal::from(100);
1110            return Ok(Value::Ratio(ratio_value, Some("percent".to_string())));
1111        }
1112
1113        // Check for "permille" keyword
1114        if peeked.kind == TokenKind::Permille {
1115            self.next()?; // consume "permille"
1116            let decimal = parse_decimal_string(num_text, &num_span, self)?;
1117            let ratio_value = decimal / Decimal::from(1000);
1118            return Ok(Value::Ratio(ratio_value, Some("permille".to_string())));
1119        }
1120
1121        // Check for duration unit
1122        if is_duration_unit(&peeked.kind) && peeked.kind != TokenKind::PercentKw {
1123            let unit_tok = self.next()?;
1124            let decimal = parse_decimal_string(num_text, &num_span, self)?;
1125            let duration_unit = token_kind_to_duration_unit(&unit_tok.kind);
1126            return Ok(Value::Duration(decimal, duration_unit));
1127        }
1128
1129        // Check for user-defined unit (identifier after number)
1130        if can_be_label(&peeked.kind) {
1131            let unit_tok = self.next()?;
1132            let decimal = parse_decimal_string(num_text, &num_span, self)?;
1133            return Ok(Value::Scale(decimal, unit_tok.text.clone()));
1134        }
1135
1136        // Plain number
1137        let decimal = parse_decimal_string(num_text, &num_span, self)?;
1138        Ok(Value::Number(decimal))
1139    }
1140
1141    fn parse_date_literal(&mut self, year_text: String, start_span: Span) -> Result<Value, Error> {
1142        let mut dt_str = year_text;
1143
1144        // Consume -MM
1145        self.expect(&TokenKind::Minus)?;
1146        dt_str.push('-');
1147        let month_tok = self.expect(&TokenKind::NumberLit)?;
1148        dt_str.push_str(&month_tok.text);
1149
1150        // Consume -DD
1151        self.expect(&TokenKind::Minus)?;
1152        dt_str.push('-');
1153        let day_tok = self.expect(&TokenKind::NumberLit)?;
1154        dt_str.push_str(&day_tok.text);
1155
1156        // Check for T (time component)
1157        if self.at(&TokenKind::Identifier)? {
1158            let peeked = self.peek()?;
1159            if peeked.text.len() >= 2
1160                && (peeked.text.starts_with('T') || peeked.text.starts_with('t'))
1161            {
1162                // The lexer may have tokenized T14 as a single identifier
1163                let t_tok = self.next()?;
1164                dt_str.push_str(&t_tok.text);
1165
1166                // Consume :MM
1167                if self.at(&TokenKind::Colon)? {
1168                    self.next()?;
1169                    dt_str.push(':');
1170                    let min_tok = self.next()?;
1171                    dt_str.push_str(&min_tok.text);
1172
1173                    // Consume :SS and optional fractional seconds
1174                    if self.at(&TokenKind::Colon)? {
1175                        self.next()?;
1176                        dt_str.push(':');
1177                        let sec_tok = self.next()?;
1178                        dt_str.push_str(&sec_tok.text);
1179
1180                        // Check for fractional seconds .NNNNNN
1181                        if self.at(&TokenKind::Dot)? {
1182                            self.next()?;
1183                            dt_str.push('.');
1184                            let frac_tok = self.expect(&TokenKind::NumberLit)?;
1185                            dt_str.push_str(&frac_tok.text);
1186                        }
1187                    }
1188                }
1189
1190                // Check for timezone
1191                self.try_consume_timezone(&mut dt_str)?;
1192            }
1193        }
1194
1195        if let Ok(dtv) = dt_str.parse::<crate::literals::DateTimeValue>() {
1196            return Ok(Value::Date(dtv));
1197        }
1198
1199        Err(Error::parsing(
1200            format!("Invalid date/time format: '{}'", dt_str),
1201            self.make_source(start_span),
1202            None::<String>,
1203        ))
1204    }
1205
1206    fn try_consume_timezone(&mut self, dt_str: &mut String) -> Result<(), Error> {
1207        // Z timezone
1208        if self.at(&TokenKind::Identifier)? {
1209            let peeked = self.peek()?;
1210            if peeked.text == "Z" || peeked.text == "z" {
1211                let z_tok = self.next()?;
1212                dt_str.push_str(&z_tok.text);
1213                return Ok(());
1214            }
1215        }
1216
1217        // +HH:MM or -HH:MM
1218        if self.at(&TokenKind::Plus)? || self.at(&TokenKind::Minus)? {
1219            let sign_tok = self.next()?;
1220            dt_str.push_str(&sign_tok.text);
1221            let hour_tok = self.expect(&TokenKind::NumberLit)?;
1222            dt_str.push_str(&hour_tok.text);
1223            if self.at(&TokenKind::Colon)? {
1224                self.next()?;
1225                dt_str.push(':');
1226                let min_tok = self.expect(&TokenKind::NumberLit)?;
1227                dt_str.push_str(&min_tok.text);
1228            }
1229        }
1230
1231        Ok(())
1232    }
1233
1234    fn try_parse_time_literal(
1235        &mut self,
1236        hour_text: String,
1237        start_span: Span,
1238    ) -> Result<Value, Error> {
1239        let mut time_str = hour_text;
1240
1241        // Consume :MM
1242        self.expect(&TokenKind::Colon)?;
1243        time_str.push(':');
1244        let min_tok = self.expect(&TokenKind::NumberLit)?;
1245        time_str.push_str(&min_tok.text);
1246
1247        // Optional :SS
1248        if self.at(&TokenKind::Colon)? {
1249            self.next()?;
1250            time_str.push(':');
1251            let sec_tok = self.expect(&TokenKind::NumberLit)?;
1252            time_str.push_str(&sec_tok.text);
1253        }
1254
1255        // Try timezone
1256        self.try_consume_timezone(&mut time_str)?;
1257
1258        if let Ok(t) = time_str.parse::<chrono::NaiveTime>() {
1259            use chrono::Timelike;
1260            return Ok(Value::Time(TimeValue {
1261                hour: t.hour() as u8,
1262                minute: t.minute() as u8,
1263                second: t.second() as u8,
1264                timezone: None,
1265            }));
1266        }
1267
1268        Err(Error::parsing(
1269            format!("Invalid time format: '{}'", time_str),
1270            self.make_source(start_span),
1271            None::<String>,
1272        ))
1273    }
1274
1275    // ========================================================================
1276    // Expression parsing (Pratt parser / precedence climbing)
1277    // ========================================================================
1278
1279    fn new_expression(
1280        &mut self,
1281        kind: ExpressionKind,
1282        source: Source,
1283    ) -> Result<Expression, Error> {
1284        self.expression_count += 1;
1285        if self.expression_count > self.max_expression_count {
1286            return Err(Error::resource_limit_exceeded(
1287                "max_expression_count",
1288                self.max_expression_count.to_string(),
1289                self.expression_count.to_string(),
1290                "Split logic into multiple rules to reduce expression count",
1291                Some(source),
1292                None,
1293                None,
1294            ));
1295        }
1296        Ok(Expression::new(kind, source))
1297    }
1298
1299    fn check_depth(&mut self) -> Result<(), Error> {
1300        if let Err(actual) = self.depth_tracker.push_depth() {
1301            let span = self.peek()?.span.clone();
1302            self.depth_tracker.pop_depth();
1303            return Err(Error::resource_limit_exceeded(
1304                "max_expression_depth",
1305                self.depth_tracker.max_depth().to_string(),
1306                actual.to_string(),
1307                "Simplify nested expressions or break into separate rules",
1308                Some(self.make_source(span)),
1309                None,
1310                None,
1311            ));
1312        }
1313        Ok(())
1314    }
1315
1316    fn parse_expression(&mut self) -> Result<Expression, Error> {
1317        self.check_depth()?;
1318        let result = self.parse_and_expression();
1319        self.depth_tracker.pop_depth();
1320        result
1321    }
1322
1323    fn parse_and_expression(&mut self) -> Result<Expression, Error> {
1324        let start_span = self.peek()?.span.clone();
1325        let mut left = self.parse_and_operand()?;
1326
1327        while self.at(&TokenKind::And)? {
1328            self.next()?; // consume 'and'
1329            let right = self.parse_and_operand()?;
1330            let span = self.span_covering(
1331                &start_span,
1332                &right
1333                    .source_location
1334                    .as_ref()
1335                    .map(|s| s.span.clone())
1336                    .unwrap_or_else(|| start_span.clone()),
1337            );
1338            left = self.new_expression(
1339                ExpressionKind::LogicalAnd(Arc::new(left), Arc::new(right)),
1340                self.make_source(span),
1341            )?;
1342        }
1343
1344        Ok(left)
1345    }
1346
1347    fn parse_and_operand(&mut self) -> Result<Expression, Error> {
1348        // not expression
1349        if self.at(&TokenKind::Not)? {
1350            return self.parse_not_expression();
1351        }
1352
1353        // base_with_suffix: base_expression followed by optional suffix
1354        self.parse_base_with_suffix()
1355    }
1356
1357    fn parse_not_expression(&mut self) -> Result<Expression, Error> {
1358        let not_tok = self.expect(&TokenKind::Not)?;
1359        let start_span = not_tok.span.clone();
1360
1361        self.check_depth()?;
1362        let operand = self.parse_and_operand()?;
1363        self.depth_tracker.pop_depth();
1364
1365        let end_span = operand
1366            .source_location
1367            .as_ref()
1368            .map(|s| s.span.clone())
1369            .unwrap_or_else(|| start_span.clone());
1370        let span = self.span_covering(&start_span, &end_span);
1371
1372        self.new_expression(
1373            ExpressionKind::LogicalNegation(Arc::new(operand), NegationType::Not),
1374            self.make_source(span),
1375        )
1376    }
1377
1378    fn parse_base_with_suffix(&mut self) -> Result<Expression, Error> {
1379        let start_span = self.peek()?.span.clone();
1380        let base = self.parse_base_expression()?;
1381
1382        // Check for suffixes
1383        let peeked = self.peek()?;
1384
1385        // Comparison suffix: >, <, >=, <=, ==, !=, is, is not
1386        if is_comparison_operator(&peeked.kind) {
1387            return self.parse_comparison_suffix(base, start_span);
1388        }
1389
1390        // "not in calendar <unit>" suffix: expr not in calendar year|month|week
1391        // After a base_expression, "not" must be this suffix (prefix "not" is only
1392        // at and_operand level, and "X and not Y" would have consumed "and" first).
1393        if peeked.kind == TokenKind::Not {
1394            return self.parse_not_in_calendar_suffix(base, start_span);
1395        }
1396
1397        // "in" suffix: conversion, date relative, date calendar
1398        if peeked.kind == TokenKind::In {
1399            return self.parse_in_suffix(base, start_span);
1400        }
1401
1402        Ok(base)
1403    }
1404
1405    fn parse_comparison_suffix(
1406        &mut self,
1407        left: Expression,
1408        start_span: Span,
1409    ) -> Result<Expression, Error> {
1410        let operator = self.parse_comparison_operator()?;
1411
1412        // Right side can be: not_expr | base_expression (optionally with "in unit")
1413        let right = if self.at(&TokenKind::Not)? {
1414            self.parse_not_expression()?
1415        } else {
1416            let rhs = self.parse_base_expression()?;
1417            // Check for "in unit" conversion on the rhs
1418            if self.at(&TokenKind::In)? {
1419                self.parse_in_suffix(rhs, start_span.clone())?
1420            } else {
1421                rhs
1422            }
1423        };
1424
1425        let end_span = right
1426            .source_location
1427            .as_ref()
1428            .map(|s| s.span.clone())
1429            .unwrap_or_else(|| start_span.clone());
1430        let span = self.span_covering(&start_span, &end_span);
1431
1432        self.new_expression(
1433            ExpressionKind::Comparison(Arc::new(left), operator, Arc::new(right)),
1434            self.make_source(span),
1435        )
1436    }
1437
1438    fn parse_comparison_operator(&mut self) -> Result<ComparisonComputation, Error> {
1439        let tok = self.next()?;
1440        match tok.kind {
1441            TokenKind::Gt => Ok(ComparisonComputation::GreaterThan),
1442            TokenKind::Lt => Ok(ComparisonComputation::LessThan),
1443            TokenKind::Gte => Ok(ComparisonComputation::GreaterThanOrEqual),
1444            TokenKind::Lte => Ok(ComparisonComputation::LessThanOrEqual),
1445            TokenKind::EqEq => Ok(ComparisonComputation::Equal),
1446            TokenKind::BangEq => Ok(ComparisonComputation::NotEqual),
1447            TokenKind::Is => {
1448                // Check for "is not"
1449                if self.at(&TokenKind::Not)? {
1450                    self.next()?; // consume 'not'
1451                    Ok(ComparisonComputation::IsNot)
1452                } else {
1453                    Ok(ComparisonComputation::Is)
1454                }
1455            }
1456            _ => Err(self.error_at_token(
1457                &tok,
1458                format!("Expected a comparison operator, found {}", tok.kind),
1459            )),
1460        }
1461    }
1462
1463    fn parse_not_in_calendar_suffix(
1464        &mut self,
1465        base: Expression,
1466        start_span: Span,
1467    ) -> Result<Expression, Error> {
1468        self.expect(&TokenKind::Not)?;
1469        self.expect(&TokenKind::In)?;
1470        self.expect(&TokenKind::Calendar)?;
1471        let unit = self.parse_calendar_unit()?;
1472        let end = self.peek()?.span.clone();
1473        let span = self.span_covering(&start_span, &end);
1474        self.new_expression(
1475            ExpressionKind::DateCalendar(DateCalendarKind::NotIn, unit, Arc::new(base)),
1476            self.make_source(span),
1477        )
1478    }
1479
1480    fn parse_in_suffix(&mut self, base: Expression, start_span: Span) -> Result<Expression, Error> {
1481        self.expect(&TokenKind::In)?;
1482
1483        let peeked = self.peek()?;
1484
1485        // "in past calendar <unit>" or "in future calendar <unit>"
1486        if peeked.kind == TokenKind::Past || peeked.kind == TokenKind::Future {
1487            let direction = self.next()?;
1488            let rel_kind = if direction.kind == TokenKind::Past {
1489                DateRelativeKind::InPast
1490            } else {
1491                DateRelativeKind::InFuture
1492            };
1493
1494            // Check for "calendar" keyword
1495            if self.at(&TokenKind::Calendar)? {
1496                self.next()?; // consume "calendar"
1497                let cal_kind = if direction.kind == TokenKind::Past {
1498                    DateCalendarKind::Past
1499                } else {
1500                    DateCalendarKind::Future
1501                };
1502                let unit = self.parse_calendar_unit()?;
1503                let end = self.peek()?.span.clone();
1504                let span = self.span_covering(&start_span, &end);
1505                return self.new_expression(
1506                    ExpressionKind::DateCalendar(cal_kind, unit, Arc::new(base)),
1507                    self.make_source(span),
1508                );
1509            }
1510
1511            // "in past [tolerance]" or "in future [tolerance]"
1512            let tolerance = if !self.at(&TokenKind::And)?
1513                && !self.at(&TokenKind::Unless)?
1514                && !self.at(&TokenKind::Then)?
1515                && !self.at(&TokenKind::Eof)?
1516                && !is_comparison_operator(&self.peek()?.kind)
1517            {
1518                let peek_kind = self.peek()?.kind.clone();
1519                if peek_kind == TokenKind::NumberLit
1520                    || peek_kind == TokenKind::LParen
1521                    || can_be_reference_segment(&peek_kind)
1522                    || is_math_function(&peek_kind)
1523                {
1524                    Some(Arc::new(self.parse_base_expression()?))
1525                } else {
1526                    None
1527                }
1528            } else {
1529                None
1530            };
1531
1532            let end = self.peek()?.span.clone();
1533            let span = self.span_covering(&start_span, &end);
1534            return self.new_expression(
1535                ExpressionKind::DateRelative(rel_kind, Arc::new(base), tolerance),
1536                self.make_source(span),
1537            );
1538        }
1539
1540        // "in calendar <unit>"
1541        if peeked.kind == TokenKind::Calendar {
1542            self.next()?; // consume "calendar"
1543            let unit = self.parse_calendar_unit()?;
1544            let end = self.peek()?.span.clone();
1545            let span = self.span_covering(&start_span, &end);
1546            return self.new_expression(
1547                ExpressionKind::DateCalendar(DateCalendarKind::Current, unit, Arc::new(base)),
1548                self.make_source(span),
1549            );
1550        }
1551
1552        // "in <unit>" — unit conversion
1553        let target_tok = self.next()?;
1554        let target = conversion_target_from_token(&target_tok.kind, &target_tok.text);
1555
1556        let converted = self.new_expression(
1557            ExpressionKind::UnitConversion(Arc::new(base), target),
1558            self.make_source(self.span_covering(&start_span, &target_tok.span)),
1559        )?;
1560
1561        // Check if followed by comparison operator
1562        if is_comparison_operator(&self.peek()?.kind) {
1563            return self.parse_comparison_suffix(converted, start_span);
1564        }
1565
1566        Ok(converted)
1567    }
1568
1569    fn parse_calendar_unit(&mut self) -> Result<CalendarUnit, Error> {
1570        let tok = self.next()?;
1571        if !is_calendar_unit_token(&tok.kind) {
1572            return Err(self.error_at_token(
1573                &tok,
1574                format!("Expected 'year', 'month', or 'week', found '{}'", tok.text),
1575            ));
1576        }
1577        Ok(token_kind_to_calendar_unit(&tok.kind))
1578    }
1579
1580    // ========================================================================
1581    // Arithmetic expressions (precedence climbing)
1582    // ========================================================================
1583
1584    fn parse_base_expression(&mut self) -> Result<Expression, Error> {
1585        let start_span = self.peek()?.span.clone();
1586        let mut left = self.parse_term()?;
1587
1588        while self.at_any(&[TokenKind::Plus, TokenKind::Minus])? {
1589            // Check if this minus is really a binary operator or could be part of something else
1590            // In "X not in calendar year", we don't want to consume "not" as an operator
1591            let op_tok = self.next()?;
1592            let operation = match op_tok.kind {
1593                TokenKind::Plus => ArithmeticComputation::Add,
1594                TokenKind::Minus => ArithmeticComputation::Subtract,
1595                _ => unreachable!("BUG: only + and - should reach here"),
1596            };
1597
1598            let right = self.parse_term()?;
1599            let end_span = right
1600                .source_location
1601                .as_ref()
1602                .map(|s| s.span.clone())
1603                .unwrap_or_else(|| start_span.clone());
1604            let span = self.span_covering(&start_span, &end_span);
1605
1606            left = self.new_expression(
1607                ExpressionKind::Arithmetic(Arc::new(left), operation, Arc::new(right)),
1608                self.make_source(span),
1609            )?;
1610        }
1611
1612        Ok(left)
1613    }
1614
1615    fn parse_term(&mut self) -> Result<Expression, Error> {
1616        let start_span = self.peek()?.span.clone();
1617        let mut left = self.parse_power()?;
1618
1619        while self.at_any(&[TokenKind::Star, TokenKind::Slash, TokenKind::Percent])? {
1620            // Be careful: % could be a percent literal suffix (e.g. 50%)
1621            // But here in term context, it's modulo since we already parsed the number
1622            let op_tok = self.next()?;
1623            let operation = match op_tok.kind {
1624                TokenKind::Star => ArithmeticComputation::Multiply,
1625                TokenKind::Slash => ArithmeticComputation::Divide,
1626                TokenKind::Percent => ArithmeticComputation::Modulo,
1627                _ => unreachable!("BUG: only *, /, % should reach here"),
1628            };
1629
1630            let right = self.parse_power()?;
1631            let end_span = right
1632                .source_location
1633                .as_ref()
1634                .map(|s| s.span.clone())
1635                .unwrap_or_else(|| start_span.clone());
1636            let span = self.span_covering(&start_span, &end_span);
1637
1638            left = self.new_expression(
1639                ExpressionKind::Arithmetic(Arc::new(left), operation, Arc::new(right)),
1640                self.make_source(span),
1641            )?;
1642        }
1643
1644        Ok(left)
1645    }
1646
1647    fn parse_power(&mut self) -> Result<Expression, Error> {
1648        let start_span = self.peek()?.span.clone();
1649        let left = self.parse_factor()?;
1650
1651        if self.at(&TokenKind::Caret)? {
1652            self.next()?;
1653            self.check_depth()?;
1654            let right = self.parse_power()?;
1655            self.depth_tracker.pop_depth();
1656            let end_span = right
1657                .source_location
1658                .as_ref()
1659                .map(|s| s.span.clone())
1660                .unwrap_or_else(|| start_span.clone());
1661            let span = self.span_covering(&start_span, &end_span);
1662
1663            return self.new_expression(
1664                ExpressionKind::Arithmetic(
1665                    Arc::new(left),
1666                    ArithmeticComputation::Power,
1667                    Arc::new(right),
1668                ),
1669                self.make_source(span),
1670            );
1671        }
1672
1673        Ok(left)
1674    }
1675
1676    fn parse_factor(&mut self) -> Result<Expression, Error> {
1677        let peeked = self.peek()?;
1678        let start_span = peeked.span.clone();
1679
1680        if peeked.kind == TokenKind::Minus {
1681            self.next()?;
1682            let operand = self.parse_primary_or_math()?;
1683            let end_span = operand
1684                .source_location
1685                .as_ref()
1686                .map(|s| s.span.clone())
1687                .unwrap_or_else(|| start_span.clone());
1688            let span = self.span_covering(&start_span, &end_span);
1689
1690            let zero = self.new_expression(
1691                ExpressionKind::Literal(Value::Number(Decimal::ZERO)),
1692                self.make_source(start_span),
1693            )?;
1694            return self.new_expression(
1695                ExpressionKind::Arithmetic(
1696                    Arc::new(zero),
1697                    ArithmeticComputation::Subtract,
1698                    Arc::new(operand),
1699                ),
1700                self.make_source(span),
1701            );
1702        }
1703
1704        if peeked.kind == TokenKind::Plus {
1705            self.next()?;
1706            return self.parse_primary_or_math();
1707        }
1708
1709        self.parse_primary_or_math()
1710    }
1711
1712    fn parse_primary_or_math(&mut self) -> Result<Expression, Error> {
1713        let peeked = self.peek()?;
1714
1715        // Math functions
1716        if is_math_function(&peeked.kind) {
1717            return self.parse_math_function();
1718        }
1719
1720        self.parse_primary()
1721    }
1722
1723    fn parse_math_function(&mut self) -> Result<Expression, Error> {
1724        let func_tok = self.next()?;
1725        let start_span = func_tok.span.clone();
1726
1727        let operator = match func_tok.kind {
1728            TokenKind::Sqrt => MathematicalComputation::Sqrt,
1729            TokenKind::Sin => MathematicalComputation::Sin,
1730            TokenKind::Cos => MathematicalComputation::Cos,
1731            TokenKind::Tan => MathematicalComputation::Tan,
1732            TokenKind::Asin => MathematicalComputation::Asin,
1733            TokenKind::Acos => MathematicalComputation::Acos,
1734            TokenKind::Atan => MathematicalComputation::Atan,
1735            TokenKind::Log => MathematicalComputation::Log,
1736            TokenKind::Exp => MathematicalComputation::Exp,
1737            TokenKind::Abs => MathematicalComputation::Abs,
1738            TokenKind::Floor => MathematicalComputation::Floor,
1739            TokenKind::Ceil => MathematicalComputation::Ceil,
1740            TokenKind::Round => MathematicalComputation::Round,
1741            _ => unreachable!("BUG: only math functions should reach here"),
1742        };
1743
1744        self.check_depth()?;
1745        let operand = self.parse_base_expression()?;
1746        self.depth_tracker.pop_depth();
1747
1748        let end_span = operand
1749            .source_location
1750            .as_ref()
1751            .map(|s| s.span.clone())
1752            .unwrap_or_else(|| start_span.clone());
1753        let span = self.span_covering(&start_span, &end_span);
1754
1755        self.new_expression(
1756            ExpressionKind::MathematicalComputation(operator, Arc::new(operand)),
1757            self.make_source(span),
1758        )
1759    }
1760
1761    fn parse_primary(&mut self) -> Result<Expression, Error> {
1762        let peeked = self.peek()?;
1763        let start_span = peeked.span.clone();
1764
1765        match &peeked.kind {
1766            // Parenthesized expression
1767            TokenKind::LParen => {
1768                self.next()?; // consume (
1769                let inner = self.parse_expression()?;
1770                self.expect(&TokenKind::RParen)?;
1771                Ok(inner)
1772            }
1773
1774            // Now keyword
1775            TokenKind::Now => {
1776                let tok = self.next()?;
1777                self.new_expression(ExpressionKind::Now, self.make_source(tok.span))
1778            }
1779
1780            // String literal
1781            TokenKind::StringLit => {
1782                let tok = self.next()?;
1783                let content = unquote_string(&tok.text);
1784                self.new_expression(
1785                    ExpressionKind::Literal(Value::Text(content)),
1786                    self.make_source(tok.span),
1787                )
1788            }
1789
1790            // Boolean literals
1791            k if is_boolean_keyword(k) => {
1792                let tok = self.next()?;
1793                self.new_expression(
1794                    ExpressionKind::Literal(Value::Boolean(token_kind_to_boolean_value(&tok.kind))),
1795                    self.make_source(tok.span),
1796                )
1797            }
1798
1799            // Number literal (could be: plain number, date, time, duration, percent, unit)
1800            TokenKind::NumberLit => self.parse_number_expression(),
1801
1802            // Reference (identifier, type keyword)
1803            k if can_be_reference_segment(k) => {
1804                let reference = self.parse_expression_reference()?;
1805                let end_span = self.peek()?.span.clone();
1806                let span = self.span_covering(&start_span, &end_span);
1807                self.new_expression(ExpressionKind::Reference(reference), self.make_source(span))
1808            }
1809
1810            _ => {
1811                let tok = self.next()?;
1812                Err(self.error_at_token(
1813                    &tok,
1814                    format!("Expected an expression, found '{}'", tok.text),
1815                ))
1816            }
1817        }
1818    }
1819
1820    fn parse_number_expression(&mut self) -> Result<Expression, Error> {
1821        let num_tok = self.next()?;
1822        let num_text = num_tok.text.clone();
1823        let start_span = num_tok.span.clone();
1824
1825        // Check if this is a date literal (YYYY-MM-DD)
1826        if num_text.len() == 4
1827            && num_text.chars().all(|c| c.is_ascii_digit())
1828            && self.at(&TokenKind::Minus)?
1829        {
1830            // Peek further: if next-next is a number, this is likely a date
1831            // We need to be careful: "2024 - 5" is arithmetic, "2024-01-15" is a date
1832            // Date format requires: YYYY-MM-DD where MM and DD are 2 digits
1833            // This is ambiguous at the token level. Let's check if the pattern matches.
1834            // Since dates use -NN- pattern and arithmetic uses - N pattern (with spaces),
1835            // we can use the span positions to disambiguate.
1836            let minus_span = self.peek()?.span.clone();
1837            // If minus is immediately adjacent to the number (no space), it's a date
1838            if minus_span.start == start_span.end {
1839                let value = self.parse_date_literal(num_text, start_span.clone())?;
1840                return self
1841                    .new_expression(ExpressionKind::Literal(value), self.make_source(start_span));
1842            }
1843        }
1844
1845        // Check for time literal (HH:MM:SS)
1846        if num_text.len() == 2
1847            && num_text.chars().all(|c| c.is_ascii_digit())
1848            && self.at(&TokenKind::Colon)?
1849        {
1850            let colon_span = self.peek()?.span.clone();
1851            if colon_span.start == start_span.end {
1852                let value = self.try_parse_time_literal(num_text, start_span.clone())?;
1853                return self
1854                    .new_expression(ExpressionKind::Literal(value), self.make_source(start_span));
1855            }
1856        }
1857
1858        // Check for %% (permille)
1859        if self.at(&TokenKind::PercentPercent)? {
1860            let pp_tok = self.next()?;
1861            if let Ok(next_peek) = self.peek() {
1862                if next_peek.kind == TokenKind::NumberLit {
1863                    return Err(self.error_at_token(
1864                        &pp_tok,
1865                        "Permille literal cannot be followed by a digit",
1866                    ));
1867                }
1868            }
1869            let decimal = parse_decimal_string(&num_text, &start_span, self)?;
1870            let ratio_value = decimal / Decimal::from(1000);
1871            return self.new_expression(
1872                ExpressionKind::Literal(Value::Ratio(ratio_value, Some("permille".to_string()))),
1873                self.make_source(start_span),
1874            );
1875        }
1876
1877        // Check for % (percent)
1878        if self.at(&TokenKind::Percent)? {
1879            let pct_span = self.peek()?.span.clone();
1880            // Only consume % if it's directly adjacent (no space) for the shorthand syntax
1881            // Or if it's "50 %" (space separated is also valid per the grammar)
1882            let pct_tok = self.next()?;
1883            if let Ok(next_peek) = self.peek() {
1884                if next_peek.kind == TokenKind::NumberLit || next_peek.kind == TokenKind::Percent {
1885                    return Err(self.error_at_token(
1886                        &pct_tok,
1887                        "Percent literal cannot be followed by a digit",
1888                    ));
1889                }
1890            }
1891            let decimal = parse_decimal_string(&num_text, &start_span, self)?;
1892            let ratio_value = decimal / Decimal::from(100);
1893            return self.new_expression(
1894                ExpressionKind::Literal(Value::Ratio(ratio_value, Some("percent".to_string()))),
1895                self.make_source(self.span_covering(&start_span, &pct_span)),
1896            );
1897        }
1898
1899        // Check for "percent" keyword
1900        if self.at(&TokenKind::PercentKw)? {
1901            self.next()?;
1902            let decimal = parse_decimal_string(&num_text, &start_span, self)?;
1903            let ratio_value = decimal / Decimal::from(100);
1904            return self.new_expression(
1905                ExpressionKind::Literal(Value::Ratio(ratio_value, Some("percent".to_string()))),
1906                self.make_source(start_span),
1907            );
1908        }
1909
1910        // Check for "permille" keyword
1911        if self.at(&TokenKind::Permille)? {
1912            self.next()?;
1913            let decimal = parse_decimal_string(&num_text, &start_span, self)?;
1914            let ratio_value = decimal / Decimal::from(1000);
1915            return self.new_expression(
1916                ExpressionKind::Literal(Value::Ratio(ratio_value, Some("permille".to_string()))),
1917                self.make_source(start_span),
1918            );
1919        }
1920
1921        // Check for duration unit
1922        if is_duration_unit(&self.peek()?.kind) && self.peek()?.kind != TokenKind::PercentKw {
1923            let unit_tok = self.next()?;
1924            let decimal = parse_decimal_string(&num_text, &start_span, self)?;
1925            let duration_unit = token_kind_to_duration_unit(&unit_tok.kind);
1926            return self.new_expression(
1927                ExpressionKind::Literal(Value::Duration(decimal, duration_unit)),
1928                self.make_source(self.span_covering(&start_span, &unit_tok.span)),
1929            );
1930        }
1931
1932        // Check for user-defined unit (identifier after number)
1933        if can_be_label(&self.peek()?.kind) {
1934            let unit_tok = self.next()?;
1935            let decimal = parse_decimal_string(&num_text, &start_span, self)?;
1936            return self.new_expression(
1937                ExpressionKind::UnresolvedUnitLiteral(decimal, unit_tok.text.clone()),
1938                self.make_source(self.span_covering(&start_span, &unit_tok.span)),
1939            );
1940        }
1941
1942        // Plain number
1943        let decimal = parse_decimal_string(&num_text, &start_span, self)?;
1944        self.new_expression(
1945            ExpressionKind::Literal(Value::Number(decimal)),
1946            self.make_source(start_span),
1947        )
1948    }
1949
1950    fn parse_expression_reference(&mut self) -> Result<Reference, Error> {
1951        let mut segments = Vec::new();
1952
1953        let first = self.next()?;
1954        segments.push(first.text.clone());
1955
1956        while self.at(&TokenKind::Dot)? {
1957            self.next()?; // consume .
1958            let seg = self.next()?;
1959            if !can_be_reference_segment(&seg.kind) {
1960                return Err(self.error_at_token(
1961                    &seg,
1962                    format!("Expected an identifier after '.', found {}", seg.kind),
1963                ));
1964            }
1965            segments.push(seg.text.clone());
1966        }
1967
1968        Ok(Reference::from_path(segments))
1969    }
1970}
1971
1972// ============================================================================
1973// Helper functions
1974// ============================================================================
1975
1976fn unquote_string(s: &str) -> String {
1977    if s.len() >= 2 && s.starts_with('"') && s.ends_with('"') {
1978        s[1..s.len() - 1].to_string()
1979    } else {
1980        s.to_string()
1981    }
1982}
1983
1984fn parse_decimal_string(text: &str, span: &Span, parser: &Parser) -> Result<Decimal, Error> {
1985    let clean = text.replace(['_', ','], "");
1986    Decimal::from_str(&clean).map_err(|_| {
1987        Error::parsing(
1988            format!(
1989                "Invalid number: '{}'. Expected a valid decimal number (e.g., 42, 3.14, 1_000_000)",
1990                text
1991            ),
1992            parser.make_source(span.clone()),
1993            None::<String>,
1994        )
1995    })
1996}
1997
1998fn is_comparison_operator(kind: &TokenKind) -> bool {
1999    matches!(
2000        kind,
2001        TokenKind::Gt
2002            | TokenKind::Lt
2003            | TokenKind::Gte
2004            | TokenKind::Lte
2005            | TokenKind::EqEq
2006            | TokenKind::BangEq
2007            | TokenKind::Is
2008    )
2009}
2010
2011// Helper trait for TokenKind
2012impl TokenKind {
2013    fn is_identifier_like(&self) -> bool {
2014        matches!(self, TokenKind::Identifier)
2015            || can_be_label(self)
2016            || is_type_keyword(self)
2017            || is_boolean_keyword(self)
2018            || is_duration_unit(self)
2019            || is_math_function(self)
2020    }
2021}