facet_solver/lib.rs
1#![cfg_attr(not(feature = "std"), no_std)]
2#![doc = include_str!("../README.md")]
3
4extern crate alloc;
5
6use alloc::borrow::Cow;
7use alloc::collections::BTreeMap;
8use alloc::collections::BTreeSet;
9use alloc::string::{String, ToString};
10use alloc::vec;
11use alloc::vec::Vec;
12use core::fmt;
13
14use facet_core::{Def, Field, Shape, StructType, Type, UserType, Variant};
15
16// Re-export resolution types from facet-reflect
17pub use facet_reflect::{
18 DuplicateFieldError, FieldCategory, FieldInfo, FieldKey, FieldPath, KeyPath, MatchResult,
19 PathSegment, Resolution, VariantSelection,
20};
21
22/// Format determines how fields are categorized and indexed in the schema.
23///
24/// Different serialization formats have different concepts of "fields":
25/// - Flat formats (JSON, TOML, YAML) treat all fields as key-value pairs
26/// - DOM formats (XML, HTML) distinguish attributes, elements, and text content
27#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
28pub enum Format {
29 /// Flat key-value formats (JSON, TOML, YAML, etc.)
30 ///
31 /// All fields are treated as keys with no distinction. The solver
32 /// uses `see_key()` to report field names.
33 #[default]
34 Flat,
35
36 /// DOM/tree formats (XML, HTML)
37 ///
38 /// Fields are categorized as attributes, elements, or text content.
39 /// The solver uses `see_attribute()`, `see_element()`, etc. to report
40 /// fields with their category.
41 Dom,
42}
43
44/// Cached schema for a type that may contain flattened fields.
45///
46/// This is computed once per Shape and can be cached forever since
47/// type information is static.
48#[derive(Debug)]
49pub struct Schema {
50 /// The shape this schema is for (kept for future caching key)
51 #[allow(dead_code)]
52 shape: &'static Shape,
53
54 /// The format this schema was built for.
55 format: Format,
56
57 /// All possible resolutions of this type.
58 /// For types with no enums in flatten paths, this has exactly 1 entry.
59 /// For types with enums, this has one entry per valid combination of variants.
60 resolutions: Vec<Resolution>,
61
62 /// Inverted index for Flat format: field_name → bitmask of configuration indices.
63 /// Bit i is set if `resolutions[i]` contains this field.
64 /// Uses a `Vec<u64>` to support arbitrary numbers of resolutions.
65 field_to_resolutions: BTreeMap<&'static str, ResolutionSet>,
66
67 /// Inverted index for Dom format: (category, name) → bitmask of configuration indices.
68 /// Only populated when format is Dom.
69 dom_field_to_resolutions: BTreeMap<(FieldCategory, &'static str), ResolutionSet>,
70}
71
72/// Handle that identifies a specific resolution inside a schema.
73#[derive(Debug, Clone, Copy)]
74pub struct ResolutionHandle<'a> {
75 index: usize,
76 resolution: &'a Resolution,
77}
78
79impl<'a> PartialEq for ResolutionHandle<'a> {
80 fn eq(&self, other: &Self) -> bool {
81 self.index == other.index
82 }
83}
84
85impl<'a> Eq for ResolutionHandle<'a> {}
86
87impl<'a> ResolutionHandle<'a> {
88 /// Internal helper to build a handle for an index within a schema.
89 fn from_schema(schema: &'a Schema, index: usize) -> Self {
90 Self {
91 index,
92 resolution: &schema.resolutions[index],
93 }
94 }
95
96 /// Resolution index within the originating schema.
97 pub const fn index(self) -> usize {
98 self.index
99 }
100
101 /// Access the underlying resolution metadata.
102 pub const fn resolution(self) -> &'a Resolution {
103 self.resolution
104 }
105}
106
107/// A set of configuration indices, stored as a bitmask for O(1) intersection.
108#[derive(Debug, Clone, PartialEq, Eq)]
109pub struct ResolutionSet {
110 /// Bitmask where bit i indicates `resolutions[i]` is in the set.
111 /// For most types, a single u64 suffices (up to 64 configs).
112 bits: Vec<u64>,
113 /// Number of resolutions in the set.
114 count: usize,
115}
116
117impl ResolutionSet {
118 /// Create an empty config set.
119 fn empty(num_resolutions: usize) -> Self {
120 let num_words = num_resolutions.div_ceil(64);
121 Self {
122 bits: vec![0; num_words],
123 count: 0,
124 }
125 }
126
127 /// Create a full config set (all configs present).
128 fn full(num_resolutions: usize) -> Self {
129 let num_words = num_resolutions.div_ceil(64);
130 let mut bits = vec![!0u64; num_words];
131 // Clear bits beyond num_resolutions
132 if !num_resolutions.is_multiple_of(64) {
133 let last_word_bits = num_resolutions % 64;
134 bits[num_words - 1] = (1u64 << last_word_bits) - 1;
135 }
136 Self {
137 bits,
138 count: num_resolutions,
139 }
140 }
141
142 /// Insert a configuration index.
143 fn insert(&mut self, idx: usize) {
144 let word = idx / 64;
145 let bit = idx % 64;
146 if self.bits[word] & (1u64 << bit) == 0 {
147 self.bits[word] |= 1u64 << bit;
148 self.count += 1;
149 }
150 }
151
152 /// Intersect with another config set in place.
153 fn intersect_with(&mut self, other: &ResolutionSet) {
154 self.count = 0;
155 for (a, b) in self.bits.iter_mut().zip(other.bits.iter()) {
156 *a &= *b;
157 self.count += a.count_ones() as usize;
158 }
159 }
160
161 /// Get the number of resolutions in the set.
162 const fn len(&self) -> usize {
163 self.count
164 }
165
166 /// Check if empty.
167 const fn is_empty(&self) -> bool {
168 self.count == 0
169 }
170
171 /// Get the first (lowest) configuration index in the set.
172 fn first(&self) -> Option<usize> {
173 for (word_idx, &word) in self.bits.iter().enumerate() {
174 if word != 0 {
175 return Some(word_idx * 64 + word.trailing_zeros() as usize);
176 }
177 }
178 None
179 }
180
181 /// Iterate over configuration indices in the set.
182 fn iter(&self) -> impl Iterator<Item = usize> + '_ {
183 self.bits.iter().enumerate().flat_map(|(word_idx, &word)| {
184 (0..64).filter_map(move |bit| {
185 if word & (1u64 << bit) != 0 {
186 Some(word_idx * 64 + bit)
187 } else {
188 None
189 }
190 })
191 })
192 }
193}
194
195/// Find fields that could disambiguate between resolutions.
196/// Returns fields that exist in some but not all resolutions.
197fn find_disambiguating_fields(configs: &[&Resolution]) -> Vec<String> {
198 if configs.len() < 2 {
199 return Vec::new();
200 }
201
202 // Collect all field names across all configs
203 let mut all_fields: BTreeSet<&str> = BTreeSet::new();
204 for config in configs {
205 for info in config.fields().values() {
206 all_fields.insert(info.serialized_name);
207 }
208 }
209
210 // Find fields that are in some but not all configs
211 let mut disambiguating = Vec::new();
212 for field in all_fields {
213 let count = configs
214 .iter()
215 .filter(|c| c.field_by_name(field).is_some())
216 .count();
217 if count > 0 && count < configs.len() {
218 disambiguating.push(field.to_string());
219 }
220 }
221
222 disambiguating
223}
224
225/// Information about a missing required field for error reporting.
226#[derive(Debug, Clone)]
227pub struct MissingFieldInfo {
228 /// The serialized field name (as it appears in input)
229 pub name: &'static str,
230 /// Full path to the field (e.g., "backend.connection.port")
231 pub path: String,
232 /// The Rust type that defines this field
233 pub defined_in: String,
234}
235
236impl MissingFieldInfo {
237 /// Create from a FieldInfo
238 fn from_field_info(info: &FieldInfo) -> Self {
239 Self {
240 name: info.serialized_name,
241 path: info.path.to_string(),
242 defined_in: info.value_shape.type_identifier.to_string(),
243 }
244 }
245}
246
247/// Information about why a specific candidate (resolution) failed to match.
248#[derive(Debug, Clone)]
249pub struct CandidateFailure {
250 /// Human-readable description of the variant (e.g., "DatabaseBackend::Postgres")
251 pub variant_name: String,
252 /// Required fields that were not provided in the input
253 pub missing_fields: Vec<MissingFieldInfo>,
254 /// Fields in the input that don't exist in this candidate
255 pub unknown_fields: Vec<String>,
256 /// Number of unknown fields that have "did you mean?" suggestions for this candidate
257 /// Higher = more likely the user intended this variant
258 pub suggestion_matches: usize,
259}
260
261/// Suggestion for a field that might have been misspelled.
262#[derive(Debug, Clone)]
263pub struct FieldSuggestion {
264 /// The unknown field from input
265 pub unknown: String,
266 /// The suggested correct field name
267 pub suggestion: &'static str,
268 /// Similarity score (0.0 to 1.0, higher is more similar)
269 pub similarity: f64,
270}
271
272/// Errors that can occur when building a schema.
273#[derive(Debug, Clone)]
274pub enum SchemaError {
275 /// A field name appears from multiple sources (parent struct and flattened struct)
276 DuplicateField(DuplicateFieldError),
277}
278
279impl From<DuplicateFieldError> for SchemaError {
280 fn from(err: DuplicateFieldError) -> Self {
281 SchemaError::DuplicateField(err)
282 }
283}
284
285impl fmt::Display for SchemaError {
286 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
287 match self {
288 SchemaError::DuplicateField(err) => {
289 write!(
290 f,
291 "Duplicate field name '{}' from different sources: {} vs {}. \
292 This usually means a parent struct and a flattened struct both \
293 define a field with the same name.",
294 err.field_name, err.first_path, err.second_path
295 )
296 }
297 }
298 }
299}
300
301#[cfg(feature = "std")]
302impl std::error::Error for SchemaError {}
303
304/// Errors that can occur during flatten resolution.
305#[derive(Debug, Clone)]
306pub enum SolverError {
307 /// No configuration matches the input fields
308 NoMatch {
309 /// The input fields that were provided
310 input_fields: Vec<String>,
311 /// Missing required fields (from the closest matching config) - simple names for backwards compat
312 missing_required: Vec<&'static str>,
313 /// Missing required fields with full path information
314 missing_required_detailed: Vec<MissingFieldInfo>,
315 /// Unknown fields that don't belong to any config
316 unknown_fields: Vec<String>,
317 /// Description of the closest matching configuration
318 closest_resolution: Option<String>,
319 /// Why each candidate failed to match (detailed per-candidate info)
320 candidate_failures: Vec<CandidateFailure>,
321 /// "Did you mean?" suggestions for unknown fields
322 suggestions: Vec<FieldSuggestion>,
323 },
324 /// Multiple resolutions match the input fields
325 Ambiguous {
326 /// Descriptions of the matching resolutions
327 candidates: Vec<String>,
328 /// Fields that could disambiguate (unique to specific configs)
329 disambiguating_fields: Vec<String>,
330 },
331}
332
333impl fmt::Display for SolverError {
334 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
335 match self {
336 SolverError::NoMatch {
337 input_fields,
338 missing_required: _,
339 missing_required_detailed,
340 unknown_fields,
341 closest_resolution,
342 candidate_failures,
343 suggestions,
344 } => {
345 write!(f, "No matching configuration for fields {input_fields:?}")?;
346
347 // Show per-candidate failure reasons if available
348 if !candidate_failures.is_empty() {
349 write!(f, "\n\nNo variant matched:")?;
350 for failure in candidate_failures {
351 write!(f, "\n - {}", failure.variant_name)?;
352 if !failure.missing_fields.is_empty() {
353 let names: Vec<_> =
354 failure.missing_fields.iter().map(|m| m.name).collect();
355 if names.len() == 1 {
356 write!(f, ": missing field '{}'", names[0])?;
357 } else {
358 write!(f, ": missing fields {names:?}")?;
359 }
360 }
361 if !failure.unknown_fields.is_empty() {
362 if failure.missing_fields.is_empty() {
363 write!(f, ":")?;
364 } else {
365 write!(f, ",")?;
366 }
367 write!(f, " unknown fields {:?}", failure.unknown_fields)?;
368 }
369 }
370 } else if let Some(config) = closest_resolution {
371 // Fallback to closest match if no per-candidate info
372 write!(f, " (closest match: {config})")?;
373 if !missing_required_detailed.is_empty() {
374 write!(f, "; missing required fields:")?;
375 for info in missing_required_detailed {
376 write!(f, " {} (at path: {})", info.name, info.path)?;
377 }
378 }
379 }
380
381 // Show unknown fields with suggestions
382 if !unknown_fields.is_empty() {
383 write!(f, "\n\nUnknown fields: {unknown_fields:?}")?;
384 }
385 for suggestion in suggestions {
386 write!(
387 f,
388 "\n Did you mean '{}' instead of '{}'?",
389 suggestion.suggestion, suggestion.unknown
390 )?;
391 }
392
393 Ok(())
394 }
395 SolverError::Ambiguous {
396 candidates,
397 disambiguating_fields,
398 } => {
399 write!(f, "Ambiguous: multiple resolutions match: {candidates:?}")?;
400 if !disambiguating_fields.is_empty() {
401 write!(
402 f,
403 "; try adding one of these fields to disambiguate: {disambiguating_fields:?}"
404 )?;
405 }
406 Ok(())
407 }
408 }
409 }
410}
411
412#[cfg(feature = "std")]
413impl std::error::Error for SolverError {}
414
415/// Compute a specificity score for a shape. Lower score = more specific.
416///
417/// This is used to disambiguate when a value could satisfy multiple types.
418/// For example, the value `42` fits both `u8` and `u16`, but `u8` is more
419/// specific (lower score), so it should be preferred.
420/// Compute a specificity score for a shape.
421///
422/// Lower score = more specific type. Used for type-based disambiguation
423/// where we want to try more specific types first (e.g., u8 before u16).
424pub fn specificity_score(shape: &'static Shape) -> u64 {
425 // Use type_identifier to determine specificity
426 // Smaller integer types are more specific
427 match shape.type_identifier {
428 "u8" | "i8" => 8,
429 "u16" | "i16" => 16,
430 "u32" | "i32" | "f32" => 32,
431 "u64" | "i64" | "f64" => 64,
432 "u128" | "i128" => 128,
433 "usize" | "isize" => 64, // Treat as 64-bit
434 // Other types get a high score (less specific)
435 _ => 1000,
436 }
437}
438
439// ============================================================================
440// Solver (State Machine)
441// ============================================================================
442
443/// Result of reporting a key to the solver.
444#[derive(Debug)]
445pub enum KeyResult<'a> {
446 /// All candidates have the same type for this key.
447 /// The deserializer can parse the value directly.
448 Unambiguous {
449 /// The shape all candidates expect for this field
450 shape: &'static Shape,
451 },
452
453 /// Candidates have different types for this key - need disambiguation.
454 /// Deserializer should parse the value, determine which fields it can
455 /// satisfy, and call `satisfy()` with the viable fields.
456 ///
457 /// **Important**: When multiple fields can be satisfied by the value,
458 /// pick the one with the lowest score (most specific). Scores are assigned
459 /// by specificity, e.g., `u8` has a lower score than `u16`.
460 Ambiguous {
461 /// The unique fields across remaining candidates (deduplicated by shape),
462 /// paired with a specificity score. Lower score = more specific type.
463 /// Deserializer should check which of these the value can satisfy,
464 /// then pick the one with the lowest score.
465 fields: Vec<(&'a FieldInfo, u64)>,
466 },
467
468 /// This key disambiguated to exactly one configuration.
469 Solved(ResolutionHandle<'a>),
470
471 /// This key doesn't exist in any remaining candidate.
472 Unknown,
473}
474
475/// Result of reporting which fields the value can satisfy.
476#[derive(Debug)]
477pub enum SatisfyResult<'a> {
478 /// Continue - still multiple candidates, keep feeding keys.
479 Continue,
480
481 /// Solved to exactly one configuration.
482 Solved(ResolutionHandle<'a>),
483
484 /// No configuration can accept the value (no fields were satisfied).
485 NoMatch,
486}
487
488/// State machine solver for lazy value-based disambiguation.
489///
490/// This solver only requests value inspection when candidates disagree on type.
491/// For keys where all candidates expect the same type, the deserializer can
492/// skip detailed value analysis.
493///
494/// # Example
495///
496/// ```rust
497/// use facet::Facet;
498/// use facet_solver::{Schema, Solver, KeyResult, SatisfyResult};
499///
500/// #[derive(Facet)]
501/// #[repr(u8)]
502/// enum NumericValue {
503/// Small(u8),
504/// Large(u16),
505/// }
506///
507/// #[derive(Facet)]
508/// struct Container {
509/// #[facet(flatten)]
510/// value: NumericValue,
511/// }
512///
513/// let schema = Schema::build(Container::SHAPE).unwrap();
514/// let mut solver = Solver::new(&schema);
515///
516/// // The field "0" has different types (u8 vs u16) - solver needs disambiguation
517/// match solver.see_key("0") {
518/// KeyResult::Ambiguous { fields } => {
519/// // Deserializer sees value "1000", checks which fields can accept it
520/// // u8 can't hold 1000, u16 can - so only report the u16 field
521/// // Fields come with specificity scores - lower = more specific
522/// let satisfied: Vec<_> = fields.iter()
523/// .filter(|(f, _score)| {
524/// // deserializer's logic: can this value parse as this field's type?
525/// f.value_shape.type_identifier == "u16"
526/// })
527/// .map(|(f, _)| *f)
528/// .collect();
529///
530/// match solver.satisfy(&satisfied) {
531/// SatisfyResult::Solved(config) => {
532/// assert!(config.resolution().describe().contains("Large"));
533/// }
534/// _ => panic!("expected solved"),
535/// }
536/// }
537/// _ => panic!("expected Ambiguous"),
538/// }
539/// ```
540#[derive(Debug)]
541pub struct Solver<'a> {
542 /// Reference to the schema for configuration lookup
543 schema: &'a Schema,
544 /// Bitmask of remaining candidate configuration indices
545 candidates: ResolutionSet,
546 /// Set of seen keys for required field checking.
547 /// For Flat format, stores FieldKey::Flat. For Dom format, stores FieldKey::Dom.
548 seen_keys: BTreeSet<FieldKey<'a>>,
549}
550
551impl<'a> Solver<'a> {
552 /// Create a new solver from a schema.
553 pub fn new(schema: &'a Schema) -> Self {
554 Self {
555 schema,
556 candidates: ResolutionSet::full(schema.resolutions.len()),
557 seen_keys: BTreeSet::new(),
558 }
559 }
560
561 /// Report a key. Returns what to do next.
562 ///
563 /// - `Unambiguous`: All candidates agree on the type - parse directly
564 /// - `Ambiguous`: Types differ - check which fields the value can satisfy
565 /// - `Solved`: Disambiguated to one config
566 /// - `Unknown`: Key not found in any candidate
567 ///
568 /// Accepts both borrowed (`&str`) and owned (`String`) keys via `Cow`.
569 /// For DOM format, use `see_attribute()`, `see_element()`, etc. instead.
570 pub fn see_key(&mut self, key: impl Into<FieldKey<'a>>) -> KeyResult<'a> {
571 let key = key.into();
572 self.see_key_internal(key)
573 }
574
575 /// Report an attribute key (DOM format only).
576 pub fn see_attribute(&mut self, name: impl Into<Cow<'a, str>>) -> KeyResult<'a> {
577 self.see_key_internal(FieldKey::attribute(name))
578 }
579
580 /// Report an element key (DOM format only).
581 pub fn see_element(&mut self, name: impl Into<Cow<'a, str>>) -> KeyResult<'a> {
582 self.see_key_internal(FieldKey::element(name))
583 }
584
585 /// Report a text content key (DOM format only).
586 pub fn see_text(&mut self) -> KeyResult<'a> {
587 self.see_key_internal(FieldKey::text())
588 }
589
590 /// Internal implementation of key lookup.
591 fn see_key_internal(&mut self, key: FieldKey<'a>) -> KeyResult<'a> {
592 self.seen_keys.insert(key.clone());
593
594 // Key-based filtering - use appropriate index based on format
595 let resolutions_with_key = match (&key, self.schema.format) {
596 (FieldKey::Flat(name), Format::Flat) => {
597 self.schema.field_to_resolutions.get(name.as_ref())
598 }
599 (FieldKey::Flat(name), Format::Dom) => {
600 // Flat key on DOM schema - try as element (most common)
601 self.schema
602 .dom_field_to_resolutions
603 .get(&(FieldCategory::Element, name.as_ref()))
604 }
605 (FieldKey::Dom(cat, name), Format::Dom) => {
606 // For Text/Tag/Elements categories, the name is often empty
607 // because there's only one such field per struct. Search by category.
608 if matches!(
609 cat,
610 FieldCategory::Text | FieldCategory::Tag | FieldCategory::Elements
611 ) && name.is_empty()
612 {
613 // Find any field with this category
614 self.schema
615 .dom_field_to_resolutions
616 .iter()
617 .find(|((c, _), _)| c == cat)
618 .map(|(_, rs)| rs)
619 } else {
620 self.schema
621 .dom_field_to_resolutions
622 .get(&(*cat, name.as_ref()))
623 }
624 }
625 (FieldKey::Dom(_, name), Format::Flat) => {
626 // DOM key on flat schema - ignore category
627 self.schema.field_to_resolutions.get(name.as_ref())
628 }
629 };
630
631 let resolutions_with_key = match resolutions_with_key {
632 Some(set) => set,
633 None => return KeyResult::Unknown,
634 };
635
636 self.candidates.intersect_with(resolutions_with_key);
637
638 if self.candidates.is_empty() {
639 return KeyResult::Unknown;
640 }
641
642 // Check if we've disambiguated to exactly one
643 if self.candidates.len() == 1 {
644 let idx = self.candidates.first().unwrap();
645 return KeyResult::Solved(self.handle(idx));
646 }
647
648 // Collect unique fields (by shape pointer) across remaining candidates
649 let mut unique_fields: Vec<&'a FieldInfo> = Vec::new();
650 for idx in self.candidates.iter() {
651 let config = &self.schema.resolutions[idx];
652 if let Some(info) = config.field_by_key(&key) {
653 // Deduplicate by shape pointer
654 if !unique_fields
655 .iter()
656 .any(|f| core::ptr::eq(f.value_shape, info.value_shape))
657 {
658 unique_fields.push(info);
659 }
660 }
661 }
662
663 if unique_fields.len() == 1 {
664 // All candidates have the same type - unambiguous
665 KeyResult::Unambiguous {
666 shape: unique_fields[0].value_shape,
667 }
668 } else if unique_fields.is_empty() {
669 KeyResult::Unknown
670 } else {
671 // Different types - need disambiguation
672 // Attach specificity scores so caller can pick most specific when multiple match
673 let fields_with_scores: Vec<_> = unique_fields
674 .into_iter()
675 .map(|f| (f, specificity_score(f.value_shape)))
676 .collect();
677 KeyResult::Ambiguous {
678 fields: fields_with_scores,
679 }
680 }
681 }
682
683 /// Report which fields the value can satisfy after `Ambiguous` result.
684 ///
685 /// The deserializer should pass the subset of fields (from the `Ambiguous` result)
686 /// that the actual value can be parsed into.
687 pub fn satisfy(&mut self, satisfied_fields: &[&FieldInfo]) -> SatisfyResult<'a> {
688 let satisfied_shapes: Vec<_> = satisfied_fields.iter().map(|f| f.value_shape).collect();
689 self.satisfy_shapes(&satisfied_shapes)
690 }
691
692 /// Report which shapes the value can satisfy after `Ambiguous` result from `probe_key`.
693 ///
694 /// This is the shape-based version of `satisfy`, used when disambiguating
695 /// by nested field types. The deserializer should pass the shapes that
696 /// the actual value can be parsed into.
697 ///
698 /// # Example
699 ///
700 /// ```rust
701 /// use facet::Facet;
702 /// use facet_solver::{Schema, Solver, KeyResult, SatisfyResult};
703 ///
704 /// #[derive(Facet)]
705 /// struct SmallPayload { value: u8 }
706 ///
707 /// #[derive(Facet)]
708 /// struct LargePayload { value: u16 }
709 ///
710 /// #[derive(Facet)]
711 /// #[repr(u8)]
712 /// enum PayloadKind {
713 /// Small { payload: SmallPayload },
714 /// Large { payload: LargePayload },
715 /// }
716 ///
717 /// #[derive(Facet)]
718 /// struct Container {
719 /// #[facet(flatten)]
720 /// inner: PayloadKind,
721 /// }
722 ///
723 /// let schema = Schema::build(Container::SHAPE).unwrap();
724 /// let mut solver = Solver::new(&schema);
725 ///
726 /// // Report nested key
727 /// solver.probe_key(&[], "payload");
728 ///
729 /// // At payload.value, value is 1000 - doesn't fit u8
730 /// // Get shapes at this path
731 /// let shapes = solver.get_shapes_at_path(&["payload", "value"]);
732 /// // Filter to shapes that can hold 1000
733 /// let works: Vec<_> = shapes.iter()
734 /// .filter(|s| s.type_identifier == "u16")
735 /// .copied()
736 /// .collect();
737 /// solver.satisfy_shapes(&works);
738 /// ```
739 pub fn satisfy_shapes(&mut self, satisfied_shapes: &[&'static Shape]) -> SatisfyResult<'a> {
740 if satisfied_shapes.is_empty() {
741 self.candidates = ResolutionSet::empty(self.schema.resolutions.len());
742 return SatisfyResult::NoMatch;
743 }
744
745 let mut new_candidates = ResolutionSet::empty(self.schema.resolutions.len());
746 for idx in self.candidates.iter() {
747 let config = &self.schema.resolutions[idx];
748 // Check if any of this config's fields match the satisfied shapes
749 for field in config.fields().values() {
750 if satisfied_shapes
751 .iter()
752 .any(|s| core::ptr::eq(*s, field.value_shape))
753 {
754 new_candidates.insert(idx);
755 break;
756 }
757 }
758 }
759 self.candidates = new_candidates;
760
761 match self.candidates.len() {
762 0 => SatisfyResult::NoMatch,
763 1 => {
764 let idx = self.candidates.first().unwrap();
765 SatisfyResult::Solved(self.handle(idx))
766 }
767 _ => SatisfyResult::Continue,
768 }
769 }
770
771 /// Get the shapes at a nested path across all remaining candidates.
772 ///
773 /// This is useful when you have an `Ambiguous` result from `probe_key`
774 /// and need to know what types are possible at that path.
775 pub fn get_shapes_at_path(&self, path: &[&str]) -> Vec<&'static Shape> {
776 let mut shapes: Vec<&'static Shape> = Vec::new();
777 for idx in self.candidates.iter() {
778 let config = &self.schema.resolutions[idx];
779 if let Some(shape) = self.get_shape_at_path(config, path)
780 && !shapes.iter().any(|s| core::ptr::eq(*s, shape))
781 {
782 shapes.push(shape);
783 }
784 }
785 shapes
786 }
787
788 /// Report which shapes at a nested path the value can satisfy.
789 ///
790 /// This is the path-aware version of `satisfy_shapes`, used when disambiguating
791 /// by nested field types after `probe_key`.
792 ///
793 /// - `path`: The full path to the field (e.g., `["payload", "value"]`)
794 /// - `satisfied_shapes`: The shapes that the value can be parsed into
795 pub fn satisfy_at_path(
796 &mut self,
797 path: &[&str],
798 satisfied_shapes: &[&'static Shape],
799 ) -> SatisfyResult<'a> {
800 if satisfied_shapes.is_empty() {
801 self.candidates = ResolutionSet::empty(self.schema.resolutions.len());
802 return SatisfyResult::NoMatch;
803 }
804
805 // Keep only candidates where the shape at this path is in the satisfied set
806 let mut new_candidates = ResolutionSet::empty(self.schema.resolutions.len());
807 for idx in self.candidates.iter() {
808 let config = &self.schema.resolutions[idx];
809 if let Some(shape) = self.get_shape_at_path(config, path)
810 && satisfied_shapes.iter().any(|s| core::ptr::eq(*s, shape))
811 {
812 new_candidates.insert(idx);
813 }
814 }
815 self.candidates = new_candidates;
816
817 match self.candidates.len() {
818 0 => SatisfyResult::NoMatch,
819 1 => {
820 let idx = self.candidates.first().unwrap();
821 SatisfyResult::Solved(self.handle(idx))
822 }
823 _ => SatisfyResult::Continue,
824 }
825 }
826
827 /// Get the current candidate resolutions.
828 pub fn candidates(&self) -> Vec<ResolutionHandle<'a>> {
829 self.candidates.iter().map(|idx| self.handle(idx)).collect()
830 }
831
832 /// Get the seen keys.
833 /// Get the seen keys.
834 pub const fn seen_keys(&self) -> &BTreeSet<FieldKey<'a>> {
835 &self.seen_keys
836 }
837
838 /// Check if a field name was seen (regardless of category for DOM format).
839 pub fn was_field_seen(&self, field_name: &str) -> bool {
840 self.seen_keys.iter().any(|k| k.name() == field_name)
841 }
842
843 #[inline]
844 fn handle(&self, idx: usize) -> ResolutionHandle<'a> {
845 ResolutionHandle::from_schema(self.schema, idx)
846 }
847
848 /// Hint that a specific enum variant should be selected.
849 ///
850 /// This filters the candidates to only those resolutions where at least one
851 /// variant selection has the given variant name. This is useful for explicit
852 /// type disambiguation via annotations (e.g., type annotations in various formats).
853 ///
854 /// Returns `true` if at least one candidate remains after filtering, `false` if
855 /// no candidates match the variant name (in which case candidates are unchanged).
856 ///
857 /// # Example
858 ///
859 /// ```rust
860 /// use facet::Facet;
861 /// use facet_solver::{Schema, Solver};
862 ///
863 /// #[derive(Facet)]
864 /// struct HttpSource { url: String }
865 ///
866 /// #[derive(Facet)]
867 /// struct GitSource { url: String, branch: String }
868 ///
869 /// #[derive(Facet)]
870 /// #[repr(u8)]
871 /// enum SourceKind {
872 /// Http(HttpSource),
873 /// Git(GitSource),
874 /// }
875 ///
876 /// #[derive(Facet)]
877 /// struct Source {
878 /// #[facet(flatten)]
879 /// kind: SourceKind,
880 /// }
881 ///
882 /// let schema = Schema::build(Source::SHAPE).unwrap();
883 /// let mut solver = Solver::new(&schema);
884 ///
885 /// // Without hint, both variants are candidates
886 /// assert_eq!(solver.candidates().len(), 2);
887 ///
888 /// // Hint at Http variant
889 /// assert!(solver.hint_variant("Http"));
890 /// assert_eq!(solver.candidates().len(), 1);
891 /// ```
892 pub fn hint_variant(&mut self, variant_name: &str) -> bool {
893 // Build a set of configs that have this variant name
894 let mut matching = ResolutionSet::empty(self.schema.resolutions.len());
895
896 for idx in self.candidates.iter() {
897 let config = &self.schema.resolutions[idx];
898 // Check if any variant selection matches the given name
899 if config
900 .variant_selections()
901 .iter()
902 .any(|vs| vs.variant_name == variant_name)
903 {
904 matching.insert(idx);
905 }
906 }
907
908 if matching.is_empty() {
909 // No matches - keep candidates unchanged
910 false
911 } else {
912 self.candidates = matching;
913 true
914 }
915 }
916
917 /// Hint that a variant is selected, but only if the field is actually a tag field
918 /// for an internally-tagged enum.
919 ///
920 /// This is safer than `hint_variant` because it validates that `tag_field_name`
921 /// is actually the tag field for an internally-tagged enum in at least one
922 /// candidate resolution before applying the hint.
923 ///
924 /// Returns `true` if the hint was applied (field was a valid tag field and
925 /// at least one candidate matches), `false` otherwise.
926 pub fn hint_variant_for_tag(&mut self, tag_field_name: &str, variant_name: &str) -> bool {
927 // First check if any candidate has this field as an internally-tagged enum tag field
928 let is_tag_field = self.candidates.iter().any(|idx| {
929 let config = &self.schema.resolutions[idx];
930 // Look for a field with the given name that is a tag field
931 config.fields().values().any(|field| {
932 field.serialized_name == tag_field_name
933 && field
934 .value_shape
935 .get_tag_attr()
936 .is_some_and(|tag| tag == tag_field_name)
937 && field.value_shape.get_content_attr().is_none()
938 })
939 });
940
941 if !is_tag_field {
942 return false;
943 }
944
945 // Now apply the variant hint
946 self.hint_variant(variant_name)
947 }
948
949 /// Mark a key as seen without filtering candidates.
950 ///
951 /// This is useful when the key is known to be present through means other than
952 /// parsing (e.g., type annotations). Call this after `hint_variant` to mark
953 /// the variant name as seen so that `finish()` doesn't report it as missing.
954 pub fn mark_seen(&mut self, key: impl Into<FieldKey<'a>>) {
955 self.seen_keys.insert(key.into());
956 }
957
958 /// Report a key at a nested path. Returns what to do next.
959 ///
960 /// This is the depth-aware version of `see_key`. Use this when probing
961 /// nested structures where disambiguation might require looking inside objects.
962 ///
963 /// - `path`: The ancestor keys (e.g., `["payload"]` when inside a payload object)
964 /// - `key`: The key found at this level (e.g., `"value"`)
965 ///
966 /// # Example
967 ///
968 /// ```rust
969 /// use facet::Facet;
970 /// use facet_solver::{Schema, Solver, KeyResult};
971 ///
972 /// #[derive(Facet)]
973 /// struct SmallPayload { value: u8 }
974 ///
975 /// #[derive(Facet)]
976 /// struct LargePayload { value: u16 }
977 ///
978 /// #[derive(Facet)]
979 /// #[repr(u8)]
980 /// enum PayloadKind {
981 /// Small { payload: SmallPayload },
982 /// Large { payload: LargePayload },
983 /// }
984 ///
985 /// #[derive(Facet)]
986 /// struct Container {
987 /// #[facet(flatten)]
988 /// inner: PayloadKind,
989 /// }
990 ///
991 /// let schema = Schema::build(Container::SHAPE).unwrap();
992 /// let mut solver = Solver::new(&schema);
993 ///
994 /// // "payload" exists in both - keep going
995 /// solver.probe_key(&[], "payload");
996 ///
997 /// // "value" inside payload - both have it but different types!
998 /// match solver.probe_key(&["payload"], "value") {
999 /// KeyResult::Ambiguous { fields } => {
1000 /// // fields is Vec<(&FieldInfo, u64)> - field + specificity score
1001 /// // Deserializer checks: 1000 fits u16 but not u8
1002 /// // When multiple match, pick the one with lowest score (most specific)
1003 /// }
1004 /// _ => {}
1005 /// }
1006 /// ```
1007 pub fn probe_key(&mut self, path: &[&str], key: &str) -> KeyResult<'a> {
1008 // Build full path
1009 let mut full_path: Vec<&str> = path.to_vec();
1010 full_path.push(key);
1011
1012 // Filter candidates to only those that have this key path
1013 let mut new_candidates = ResolutionSet::empty(self.schema.resolutions.len());
1014 for idx in self.candidates.iter() {
1015 let config = &self.schema.resolutions[idx];
1016 if config.has_key_path(&full_path) {
1017 new_candidates.insert(idx);
1018 }
1019 }
1020 self.candidates = new_candidates;
1021
1022 if self.candidates.is_empty() {
1023 return KeyResult::Unknown;
1024 }
1025
1026 // Check if we've disambiguated to exactly one
1027 if self.candidates.len() == 1 {
1028 let idx = self.candidates.first().unwrap();
1029 return KeyResult::Solved(self.handle(idx));
1030 }
1031
1032 // Get the shape at this path for each remaining candidate
1033 // We need to traverse the type tree to find the actual field type
1034 let mut unique_shapes: Vec<(&'static Shape, usize)> = Vec::new(); // (shape, resolution_idx)
1035
1036 for idx in self.candidates.iter() {
1037 let config = &self.schema.resolutions[idx];
1038 if let Some(shape) = self.get_shape_at_path(config, &full_path) {
1039 // Deduplicate by shape pointer
1040 if !unique_shapes.iter().any(|(s, _)| core::ptr::eq(*s, shape)) {
1041 unique_shapes.push((shape, idx));
1042 }
1043 }
1044 }
1045
1046 match unique_shapes.len() {
1047 0 => KeyResult::Unknown,
1048 1 => {
1049 // All candidates have the same type at this path - unambiguous
1050 KeyResult::Unambiguous {
1051 shape: unique_shapes[0].0,
1052 }
1053 }
1054 _ => {
1055 // Different types at this path - need disambiguation
1056 // Build FieldInfo with scores for each unique shape
1057 let fields: Vec<(&'a FieldInfo, u64)> = unique_shapes
1058 .iter()
1059 .filter_map(|(shape, idx)| {
1060 let config = &self.schema.resolutions[*idx];
1061 // For nested paths, we need the parent field
1062 // e.g., for ["payload", "value"], get the "payload" field
1063 let field = if path.is_empty() {
1064 config.field_by_name(key)
1065 } else {
1066 // Return the top-level field that contains this path
1067 config.field_by_name(path[0])
1068 }?;
1069 Some((field, specificity_score(shape)))
1070 })
1071 .collect();
1072
1073 KeyResult::Ambiguous { fields }
1074 }
1075 }
1076 }
1077
1078 /// Get the shape at a nested path within a configuration.
1079 fn get_shape_at_path(&self, config: &'a Resolution, path: &[&str]) -> Option<&'static Shape> {
1080 if path.is_empty() {
1081 return None;
1082 }
1083
1084 // Start with the top-level field
1085 let top_field = config.field_by_name(path[0])?;
1086 let mut current_shape = top_field.value_shape;
1087
1088 // Navigate through nested structs
1089 for &key in &path[1..] {
1090 current_shape = self.get_field_shape(current_shape, key)?;
1091 }
1092
1093 Some(current_shape)
1094 }
1095
1096 /// Get the shape of a field within a struct shape.
1097 fn get_field_shape(&self, shape: &'static Shape, field_name: &str) -> Option<&'static Shape> {
1098 use facet_core::{StructType, Type, UserType};
1099
1100 match shape.ty {
1101 Type::User(UserType::Struct(StructType { fields, .. })) => {
1102 for field in fields {
1103 if field.effective_name() == field_name {
1104 return Some(field.shape());
1105 }
1106 }
1107 None
1108 }
1109 _ => None,
1110 }
1111 }
1112
1113 /// Finish solving. Call this after all keys have been processed.
1114 ///
1115 /// This method is necessary because key-based filtering alone cannot disambiguate
1116 /// when one variant's required fields are a subset of another's.
1117 ///
1118 /// # Why not just use `see_key()` results?
1119 ///
1120 /// `see_key()` returns `Solved` when a key *excludes* candidates down to one.
1121 /// But when the input is a valid subset of multiple variants, no key excludes
1122 /// anything — you need `finish()` to check which candidates have all their
1123 /// required fields satisfied.
1124 ///
1125 /// # Example
1126 ///
1127 /// ```rust,ignore
1128 /// enum Source {
1129 /// Http { url: String }, // required: url
1130 /// Git { url: String, branch: String }, // required: url, branch
1131 /// }
1132 /// ```
1133 ///
1134 /// | Input | `see_key` behavior | Resolution |
1135 /// |------------------------|-------------------------------------------|-----------------------|
1136 /// | `{ "url", "branch" }` | `branch` excludes `Http` → candidates = 1 | Early `Solved(Git)` |
1137 /// | `{ "url" }` | both have `url` → candidates = 2 | `finish()` → `Http` |
1138 ///
1139 /// In the second case, no key ever excludes a candidate. Only `finish()` can
1140 /// determine that `Git` is missing its required `branch` field, leaving `Http`
1141 /// as the sole viable configuration.
1142 #[allow(clippy::result_large_err)] // SolverError intentionally contains detailed diagnostic info
1143 pub fn finish(self) -> Result<ResolutionHandle<'a>, SolverError> {
1144 let Solver {
1145 schema,
1146 candidates,
1147 seen_keys,
1148 } = self;
1149
1150 // Compute all known fields across all resolutions (for unknown field detection)
1151 let all_known_fields: BTreeSet<&'static str> = schema
1152 .resolutions
1153 .iter()
1154 .flat_map(|r| r.fields().values().map(|f| f.serialized_name))
1155 .collect();
1156
1157 // Find unknown fields (fields in input that don't exist in ANY resolution)
1158 let unknown_fields: Vec<String> = seen_keys
1159 .iter()
1160 .filter(|k| !all_known_fields.contains(k.name()))
1161 .map(|k| k.name().to_string())
1162 .collect();
1163
1164 // Compute suggestions for unknown fields
1165 let suggestions = compute_suggestions(&unknown_fields, &all_known_fields);
1166
1167 if candidates.is_empty() {
1168 // Build per-candidate failure info for all resolutions
1169 let mut candidate_failures: Vec<CandidateFailure> = schema
1170 .resolutions
1171 .iter()
1172 .map(|config| build_candidate_failure(config, &seen_keys))
1173 .collect();
1174
1175 // Sort by closeness (best match first)
1176 sort_candidates_by_closeness(&mut candidate_failures);
1177
1178 return Err(SolverError::NoMatch {
1179 input_fields: seen_keys.iter().map(|k| k.name().to_string()).collect(),
1180 missing_required: Vec::new(),
1181 missing_required_detailed: Vec::new(),
1182 unknown_fields,
1183 closest_resolution: None,
1184 candidate_failures,
1185 suggestions,
1186 });
1187 }
1188
1189 // Filter candidates to only those that have all required fields satisfied
1190 let viable: Vec<usize> = candidates
1191 .iter()
1192 .filter(|idx| {
1193 let config = &schema.resolutions[*idx];
1194 config
1195 .required_field_names()
1196 .iter()
1197 .all(|f| seen_keys.iter().any(|k| k.name() == *f))
1198 })
1199 .collect();
1200
1201 match viable.len() {
1202 0 => {
1203 // No viable candidates - build per-candidate failure info
1204 let mut candidate_failures: Vec<CandidateFailure> = candidates
1205 .iter()
1206 .map(|idx| {
1207 let config = &schema.resolutions[idx];
1208 build_candidate_failure(config, &seen_keys)
1209 })
1210 .collect();
1211
1212 // Sort by closeness (best match first)
1213 sort_candidates_by_closeness(&mut candidate_failures);
1214
1215 // For backwards compatibility, also populate the "closest" fields
1216 // Now use the first (closest) candidate after sorting
1217 let closest_name = candidate_failures.first().map(|f| f.variant_name.clone());
1218 let closest_config = closest_name
1219 .as_ref()
1220 .and_then(|name| schema.resolutions.iter().find(|r| r.describe() == *name));
1221
1222 let (missing, missing_detailed, closest_resolution) =
1223 if let Some(config) = closest_config {
1224 let missing: Vec<_> = config
1225 .required_field_names()
1226 .iter()
1227 .filter(|f| !seen_keys.iter().any(|k| k.name() == **f))
1228 .copied()
1229 .collect();
1230 let missing_detailed: Vec<_> = missing
1231 .iter()
1232 .filter_map(|name| config.field_by_name(name))
1233 .map(MissingFieldInfo::from_field_info)
1234 .collect();
1235 (missing, missing_detailed, Some(config.describe()))
1236 } else {
1237 (Vec::new(), Vec::new(), None)
1238 };
1239
1240 Err(SolverError::NoMatch {
1241 input_fields: seen_keys.iter().map(|s| s.to_string()).collect(),
1242 missing_required: missing,
1243 missing_required_detailed: missing_detailed,
1244 unknown_fields,
1245 closest_resolution,
1246 candidate_failures,
1247 suggestions,
1248 })
1249 }
1250 1 => {
1251 // Exactly one viable candidate - success!
1252 Ok(ResolutionHandle::from_schema(schema, viable[0]))
1253 }
1254 _ => {
1255 // Multiple viable candidates - ambiguous!
1256 let configs: Vec<_> = viable.iter().map(|&idx| &schema.resolutions[idx]).collect();
1257 let candidates: Vec<String> = configs.iter().map(|c| c.describe()).collect();
1258 let disambiguating_fields = find_disambiguating_fields(&configs);
1259
1260 Err(SolverError::Ambiguous {
1261 candidates,
1262 disambiguating_fields,
1263 })
1264 }
1265 }
1266 }
1267}
1268
1269/// Build a CandidateFailure for a resolution given the seen keys.
1270fn build_candidate_failure<'a>(
1271 config: &Resolution,
1272 seen_keys: &BTreeSet<FieldKey<'a>>,
1273) -> CandidateFailure {
1274 let missing_fields: Vec<MissingFieldInfo> = config
1275 .required_field_names()
1276 .iter()
1277 .filter(|f| !seen_keys.iter().any(|k| k.name() == **f))
1278 .filter_map(|f| config.field_by_name(f))
1279 .map(MissingFieldInfo::from_field_info)
1280 .collect();
1281
1282 let unknown_fields: Vec<String> = seen_keys
1283 .iter()
1284 .filter(|k| config.field_by_key(k).is_none())
1285 .map(|k| k.name().to_string())
1286 .collect();
1287
1288 // Compute closeness score for ranking
1289 let suggestion_matches = compute_closeness_score(&unknown_fields, &missing_fields, config);
1290
1291 CandidateFailure {
1292 variant_name: config.describe(),
1293 missing_fields,
1294 unknown_fields,
1295 suggestion_matches,
1296 }
1297}
1298
1299/// Compute a closeness score for ranking candidates.
1300/// Higher score = more likely the user intended this variant.
1301///
1302/// The score considers:
1303/// - Typo matches: unknown fields that are similar to known fields (weighted by similarity)
1304/// - Field coverage: if we fixed typos, would we have all required fields?
1305/// - Missing fields: fewer missing = better
1306/// - Unknown fields: fewer truly unknown (no suggestion) = better
1307#[cfg(feature = "suggestions")]
1308fn compute_closeness_score(
1309 unknown_fields: &[String],
1310 missing_fields: &[MissingFieldInfo],
1311 config: &Resolution,
1312) -> usize {
1313 const SIMILARITY_THRESHOLD: f64 = 0.6;
1314
1315 // Score components (scaled to integers for easy comparison)
1316 let mut typo_score: usize = 0;
1317 let mut fields_that_would_match: usize = 0;
1318
1319 // For each unknown field, find best matching known field
1320 for unknown in unknown_fields {
1321 let mut best_similarity = 0.0f64;
1322 let mut best_match: Option<&str> = None;
1323
1324 for info in config.fields().values() {
1325 let similarity = strsim::jaro_winkler(unknown, info.serialized_name);
1326 if similarity >= SIMILARITY_THRESHOLD && similarity > best_similarity {
1327 best_similarity = similarity;
1328 best_match = Some(info.serialized_name);
1329 }
1330 }
1331
1332 if let Some(_matched_field) = best_match {
1333 // Weight by similarity: 0.6 -> 60 points, 1.0 -> 100 points
1334 typo_score += (best_similarity * 100.0) as usize;
1335 fields_that_would_match += 1;
1336 }
1337 }
1338
1339 // Calculate how many required fields would be satisfied if typos were fixed
1340 let required_count = config.required_field_names().len();
1341 let currently_missing = missing_fields.len();
1342 let would_be_missing = currently_missing.saturating_sub(fields_that_would_match);
1343
1344 // Coverage score: percentage of required fields that would be present
1345 let coverage_score = if required_count > 0 {
1346 ((required_count - would_be_missing) * 100) / required_count
1347 } else {
1348 100 // No required fields = perfect coverage
1349 };
1350
1351 // Penalty for truly unknown fields (no typo suggestion)
1352 let truly_unknown = unknown_fields.len().saturating_sub(fields_that_would_match);
1353 let unknown_penalty = truly_unknown * 10;
1354
1355 // Combine scores: typo matches are most important, then coverage, then penalties
1356 // Each typo match can give up to 100 points, so scale coverage to match
1357 typo_score + coverage_score.saturating_sub(unknown_penalty)
1358}
1359
1360/// Compute closeness score (no-op without suggestions feature).
1361#[cfg(not(feature = "suggestions"))]
1362fn compute_closeness_score(
1363 _unknown_fields: &[String],
1364 _missing_fields: &[MissingFieldInfo],
1365 _config: &Resolution,
1366) -> usize {
1367 0
1368}
1369
1370/// Sort candidate failures by closeness (best match first).
1371fn sort_candidates_by_closeness(failures: &mut [CandidateFailure]) {
1372 failures.sort_by(|a, b| {
1373 // Higher suggestion_matches (closeness score) first
1374 b.suggestion_matches.cmp(&a.suggestion_matches)
1375 });
1376}
1377
1378/// Compute "did you mean?" suggestions for unknown fields.
1379#[cfg(feature = "suggestions")]
1380fn compute_suggestions(
1381 unknown_fields: &[String],
1382 all_known_fields: &BTreeSet<&'static str>,
1383) -> Vec<FieldSuggestion> {
1384 const SIMILARITY_THRESHOLD: f64 = 0.6;
1385
1386 let mut suggestions = Vec::new();
1387
1388 for unknown in unknown_fields {
1389 let mut best_match: Option<(&'static str, f64)> = None;
1390
1391 for known in all_known_fields {
1392 let similarity = strsim::jaro_winkler(unknown, known);
1393 if similarity >= SIMILARITY_THRESHOLD
1394 && best_match.is_none_or(|(_, best_sim)| similarity > best_sim)
1395 {
1396 best_match = Some((known, similarity));
1397 }
1398 }
1399
1400 if let Some((suggestion, similarity)) = best_match {
1401 suggestions.push(FieldSuggestion {
1402 unknown: unknown.clone(),
1403 suggestion,
1404 similarity,
1405 });
1406 }
1407 }
1408
1409 suggestions
1410}
1411
1412/// Compute "did you mean?" suggestions for unknown fields (no-op without strsim).
1413#[cfg(not(feature = "suggestions"))]
1414fn compute_suggestions(
1415 _unknown_fields: &[String],
1416 _all_known_fields: &BTreeSet<&'static str>,
1417) -> Vec<FieldSuggestion> {
1418 Vec::new()
1419}
1420
1421// ============================================================================
1422// Probing Solver (Depth-Aware)
1423// ============================================================================
1424
1425/// Result of reporting a key to the probing solver.
1426#[derive(Debug)]
1427pub enum ProbeResult<'a> {
1428 /// Keep reporting keys - not yet disambiguated
1429 KeepGoing,
1430 /// Solved! Use this configuration
1431 Solved(&'a Resolution),
1432 /// No configuration matches the observed keys
1433 NoMatch,
1434}
1435
1436/// Depth-aware probing solver for streaming deserialization.
1437///
1438/// Unlike the batch solver, this solver accepts
1439/// key reports at arbitrary depths. It's designed for the "peek" strategy:
1440///
1441/// 1. Deserializer scans keys (without parsing values) and reports them
1442/// 2. Solver filters candidates based on which configs have that key path
1443/// 3. Once one candidate remains, solver returns `Solved`
1444/// 4. Deserializer rewinds and parses into the resolved type
1445///
1446/// # Example
1447///
1448/// ```rust
1449/// use facet::Facet;
1450/// use facet_solver::{Schema, ProbingSolver, ProbeResult};
1451///
1452/// #[derive(Facet)]
1453/// struct TextPayload { content: String }
1454///
1455/// #[derive(Facet)]
1456/// struct BinaryPayload { bytes: Vec<u8> }
1457///
1458/// #[derive(Facet)]
1459/// #[repr(u8)]
1460/// enum MessageKind {
1461/// Text { payload: TextPayload },
1462/// Binary { payload: BinaryPayload },
1463/// }
1464///
1465/// #[derive(Facet)]
1466/// struct Message {
1467/// id: String,
1468/// #[facet(flatten)]
1469/// kind: MessageKind,
1470/// }
1471///
1472/// let schema = Schema::build(Message::SHAPE).unwrap();
1473/// let mut solver = ProbingSolver::new(&schema);
1474///
1475/// // "id" exists in both configs - keep going
1476/// assert!(matches!(solver.probe_key(&[], "id"), ProbeResult::KeepGoing));
1477///
1478/// // "payload" exists in both configs - keep going
1479/// assert!(matches!(solver.probe_key(&[], "payload"), ProbeResult::KeepGoing));
1480///
1481/// // "content" inside payload only exists in Text - solved!
1482/// match solver.probe_key(&["payload"], "content") {
1483/// ProbeResult::Solved(config) => {
1484/// assert!(config.has_key_path(&["payload", "content"]));
1485/// }
1486/// _ => panic!("expected Solved"),
1487/// }
1488/// ```
1489#[derive(Debug)]
1490pub struct ProbingSolver<'a> {
1491 /// Remaining candidate resolutions
1492 candidates: Vec<&'a Resolution>,
1493}
1494
1495impl<'a> ProbingSolver<'a> {
1496 /// Create a new probing solver from a schema.
1497 pub fn new(schema: &'a Schema) -> Self {
1498 Self {
1499 candidates: schema.resolutions.iter().collect(),
1500 }
1501 }
1502
1503 /// Create a new probing solver from resolutions directly.
1504 pub fn from_resolutions(configs: &'a [Resolution]) -> Self {
1505 Self {
1506 candidates: configs.iter().collect(),
1507 }
1508 }
1509
1510 /// Report a key found at a path during probing.
1511 ///
1512 /// - `path`: The ancestor keys (e.g., `["payload"]` when inside the payload object)
1513 /// - `key`: The key found at this level (e.g., `"content"`)
1514 ///
1515 /// Returns what to do next.
1516 pub fn probe_key(&mut self, path: &[&str], key: &str) -> ProbeResult<'a> {
1517 // Build the full key path (runtime strings, compared against static schema)
1518 let mut full_path: Vec<&str> = path.to_vec();
1519 full_path.push(key);
1520
1521 // Filter to candidates that have this key path
1522 self.candidates.retain(|c| c.has_key_path(&full_path));
1523
1524 match self.candidates.len() {
1525 0 => ProbeResult::NoMatch,
1526 1 => ProbeResult::Solved(self.candidates[0]),
1527 _ => ProbeResult::KeepGoing,
1528 }
1529 }
1530
1531 /// Get the current candidate resolutions.
1532 pub fn candidates(&self) -> &[&'a Resolution] {
1533 &self.candidates
1534 }
1535
1536 /// Finish probing - returns Solved if exactly one candidate remains.
1537 pub fn finish(&self) -> ProbeResult<'a> {
1538 match self.candidates.len() {
1539 0 => ProbeResult::NoMatch,
1540 1 => ProbeResult::Solved(self.candidates[0]),
1541 _ => ProbeResult::KeepGoing, // Still ambiguous
1542 }
1543 }
1544}
1545
1546// ============================================================================
1547// Variant Format Classification
1548// ============================================================================
1549
1550/// Classification of an enum variant's expected serialized format.
1551///
1552/// This is used by deserializers to determine how to parse untagged enum variants
1553/// based on the YAML/JSON/etc. value type they encounter.
1554#[derive(Debug, Clone, Copy, PartialEq, Eq)]
1555pub enum VariantFormat {
1556 /// Unit variant: no fields, serializes as the variant name or nothing for untagged
1557 Unit,
1558
1559 /// Newtype variant wrapping a scalar type (String, numbers, bool, etc.)
1560 /// Serializes as just the scalar value for untagged enums.
1561 NewtypeScalar {
1562 /// The shape of the inner scalar type
1563 inner_shape: &'static Shape,
1564 },
1565
1566 /// Newtype variant wrapping a struct
1567 /// Serializes as a mapping for untagged enums.
1568 NewtypeStruct {
1569 /// The shape of the inner struct type
1570 inner_shape: &'static Shape,
1571 },
1572
1573 /// Newtype variant wrapping a tuple struct/tuple
1574 /// Serializes as a sequence for untagged enums.
1575 NewtypeTuple {
1576 /// The shape of the inner tuple type
1577 inner_shape: &'static Shape,
1578 /// Number of elements in the inner tuple
1579 arity: usize,
1580 },
1581
1582 /// Newtype variant wrapping a sequence type (Vec, Array, Slice, Set)
1583 /// Serializes as a sequence for untagged enums.
1584 NewtypeSequence {
1585 /// The shape of the inner sequence type
1586 inner_shape: &'static Shape,
1587 },
1588
1589 /// Newtype variant wrapping another type (enum, map, etc.)
1590 NewtypeOther {
1591 /// The shape of the inner type
1592 inner_shape: &'static Shape,
1593 },
1594
1595 /// Tuple variant with multiple fields
1596 /// Serializes as a sequence for untagged enums.
1597 Tuple {
1598 /// Number of fields in the tuple
1599 arity: usize,
1600 },
1601
1602 /// Struct variant with named fields
1603 /// Serializes as a mapping for untagged enums.
1604 Struct,
1605}
1606
1607impl VariantFormat {
1608 /// Classify a variant's expected serialized format.
1609 pub fn from_variant(variant: &'static Variant) -> Self {
1610 use facet_core::StructKind;
1611
1612 let fields = variant.data.fields;
1613 let kind = variant.data.kind;
1614
1615 match kind {
1616 StructKind::Unit => VariantFormat::Unit,
1617 // TupleStruct and Tuple are both used for tuple-like variants
1618 // depending on how they're defined. Handle them the same way.
1619 StructKind::TupleStruct | StructKind::Tuple => {
1620 if fields.len() == 1 {
1621 // Newtype variant - classify by inner type
1622 let field_shape = fields[0].shape();
1623 // Dereference through pointers to get the actual inner type
1624 let inner_shape = deref_pointer(field_shape);
1625
1626 // Check if this is a Spanned<T> wrapper and unwrap it for classification
1627 // This allows untagged enum variants containing Spanned<String> etc.
1628 // to match scalar values transparently
1629 let classification_shape = if let Some(spanned_inner) =
1630 facet_reflect::get_spanned_inner_shape(field_shape)
1631 {
1632 spanned_inner
1633 } else {
1634 field_shape
1635 };
1636
1637 if is_scalar_shape(classification_shape)
1638 || is_unit_enum_shape(classification_shape)
1639 {
1640 // Scalars and unit-only enums both serialize as primitive values
1641 // Store the classification shape (unwrapped from Spanned if needed)
1642 // so that type matching works correctly for multi-variant untagged enums
1643 VariantFormat::NewtypeScalar {
1644 inner_shape: classification_shape,
1645 }
1646 } else if let Some(arity) = tuple_struct_arity(classification_shape) {
1647 VariantFormat::NewtypeTuple { inner_shape, arity }
1648 } else if is_named_struct_shape(classification_shape) {
1649 VariantFormat::NewtypeStruct { inner_shape }
1650 } else if is_sequence_shape(classification_shape) {
1651 VariantFormat::NewtypeSequence { inner_shape }
1652 } else {
1653 VariantFormat::NewtypeOther { inner_shape }
1654 }
1655 } else {
1656 // Multi-field tuple variant
1657 VariantFormat::Tuple {
1658 arity: fields.len(),
1659 }
1660 }
1661 }
1662 StructKind::Struct => VariantFormat::Struct,
1663 }
1664 }
1665
1666 /// Returns true if this variant expects a scalar value in untagged format.
1667 pub const fn expects_scalar(&self) -> bool {
1668 matches!(self, VariantFormat::NewtypeScalar { .. })
1669 }
1670
1671 /// Returns true if this variant expects a sequence in untagged format.
1672 pub const fn expects_sequence(&self) -> bool {
1673 matches!(
1674 self,
1675 VariantFormat::Tuple { .. }
1676 | VariantFormat::NewtypeTuple { .. }
1677 | VariantFormat::NewtypeSequence { .. }
1678 )
1679 }
1680
1681 /// Returns true if this variant expects a mapping in untagged format.
1682 pub const fn expects_mapping(&self) -> bool {
1683 matches!(
1684 self,
1685 VariantFormat::Struct | VariantFormat::NewtypeStruct { .. }
1686 )
1687 }
1688
1689 /// Returns true if this is a unit variant (no data).
1690 pub const fn is_unit(&self) -> bool {
1691 matches!(self, VariantFormat::Unit)
1692 }
1693}
1694
1695/// Dereference through pointer types (like `Box<T>`) to get the pointee shape.
1696/// Returns the original shape if it's not a pointer.
1697fn deref_pointer(shape: &'static Shape) -> &'static Shape {
1698 use facet_core::Def;
1699
1700 match shape.def {
1701 Def::Pointer(pointer_def) => {
1702 if let Some(pointee) = pointer_def.pointee() {
1703 // Recursively dereference in case of nested pointers
1704 deref_pointer(pointee)
1705 } else {
1706 // Opaque pointer - can't dereference
1707 shape
1708 }
1709 }
1710 _ => shape,
1711 }
1712}
1713
1714/// Check if a shape represents a scalar type.
1715/// Transparently handles pointer types like `Box<i32>`.
1716fn is_scalar_shape(shape: &'static Shape) -> bool {
1717 let shape = deref_pointer(shape);
1718 shape.scalar_type().is_some()
1719}
1720
1721/// Returns the arity of a tuple struct/tuple shape, if applicable.
1722/// Transparently handles pointer types like `Box<(i32, i32)>`.
1723fn tuple_struct_arity(shape: &'static Shape) -> Option<usize> {
1724 use facet_core::{StructKind, Type, UserType};
1725
1726 let shape = deref_pointer(shape);
1727 match shape.ty {
1728 Type::User(UserType::Struct(struct_type)) => match struct_type.kind {
1729 StructKind::Tuple | StructKind::TupleStruct => Some(struct_type.fields.len()),
1730 _ => None,
1731 },
1732 _ => None,
1733 }
1734}
1735
1736/// Returns true if the shape is a named struct (non-tuple).
1737/// Transparently handles pointer types like `Box<MyStruct>`.
1738fn is_named_struct_shape(shape: &'static Shape) -> bool {
1739 use facet_core::{StructKind, Type, UserType};
1740
1741 let shape = deref_pointer(shape);
1742 matches!(
1743 shape.ty,
1744 Type::User(UserType::Struct(struct_type)) if matches!(struct_type.kind, StructKind::Struct)
1745 )
1746}
1747
1748/// Returns true if the shape is a sequence type (List, Array, Slice, Set).
1749/// These types serialize as arrays/sequences in formats like TOML, JSON, YAML.
1750/// Transparently handles pointer types like `Box<Vec<i32>>`.
1751fn is_sequence_shape(shape: &'static Shape) -> bool {
1752 use facet_core::Def;
1753
1754 let shape = deref_pointer(shape);
1755 matches!(
1756 shape.def,
1757 Def::List(_) | Def::Array(_) | Def::Slice(_) | Def::Set(_)
1758 )
1759}
1760
1761/// Returns true if the shape is a unit-only enum.
1762/// Unit-only enums serialize as strings in most formats (TOML, JSON, YAML).
1763/// Transparently handles pointer types like `Box<UnitEnum>`.
1764fn is_unit_enum_shape(shape: &'static Shape) -> bool {
1765 use facet_core::{Type, UserType};
1766
1767 let shape = deref_pointer(shape);
1768 match shape.ty {
1769 Type::User(UserType::Enum(enum_type)) => {
1770 // Check if all variants are unit variants
1771 enum_type.variants.iter().all(|v| v.data.fields.is_empty())
1772 }
1773 _ => false,
1774 }
1775}
1776
1777/// Information about variants grouped by their expected format.
1778///
1779/// Used by deserializers to efficiently dispatch untagged enum parsing
1780/// based on the type of value encountered.
1781#[derive(Debug, Default)]
1782pub struct VariantsByFormat {
1783 /// Variants that expect a scalar value (newtype wrapping String, i32, etc.)
1784 ///
1785 /// **Deprecated:** Use the type-specific fields below for better type matching.
1786 /// This field contains all scalar variants regardless of type.
1787 pub scalar_variants: Vec<(&'static Variant, &'static Shape)>,
1788
1789 /// Variants that expect a boolean value (newtype wrapping bool)
1790 pub bool_variants: Vec<(&'static Variant, &'static Shape)>,
1791
1792 /// Variants that expect an integer value (newtype wrapping i8, u8, i32, u64, etc.)
1793 pub int_variants: Vec<(&'static Variant, &'static Shape)>,
1794
1795 /// Variants that expect a float value (newtype wrapping f32, f64)
1796 pub float_variants: Vec<(&'static Variant, &'static Shape)>,
1797
1798 /// Variants that expect a string value (newtype wrapping String, `&str`, `Cow<str>`)
1799 pub string_variants: Vec<(&'static Variant, &'static Shape)>,
1800
1801 /// Variants that expect a sequence (tuple variants)
1802 /// Grouped by arity for efficient matching.
1803 pub tuple_variants: Vec<(&'static Variant, usize)>,
1804
1805 /// Variants that expect a mapping (struct variants, newtype wrapping struct)
1806 pub struct_variants: Vec<&'static Variant>,
1807
1808 /// Unit variants (no data)
1809 pub unit_variants: Vec<&'static Variant>,
1810
1811 /// Other variants that don't fit the above categories
1812 pub other_variants: Vec<&'static Variant>,
1813}
1814
1815impl VariantsByFormat {
1816 /// Build variant classification for an enum shape.
1817 ///
1818 /// Returns None if the shape is not an enum.
1819 pub fn from_shape(shape: &'static Shape) -> Option<Self> {
1820 use facet_core::{Type, UserType};
1821
1822 let enum_type = match shape.ty {
1823 Type::User(UserType::Enum(e)) => e,
1824 _ => return None,
1825 };
1826
1827 let mut result = Self::default();
1828
1829 for variant in enum_type.variants {
1830 match VariantFormat::from_variant(variant) {
1831 VariantFormat::Unit => {
1832 result.unit_variants.push(variant);
1833 }
1834 VariantFormat::NewtypeScalar { inner_shape } => {
1835 // Add to general scalar_variants (for backward compatibility)
1836 result.scalar_variants.push((variant, inner_shape));
1837
1838 // Classify by specific scalar type for better type matching
1839 // Dereference through pointers (Box, &, etc.) to get the actual scalar type
1840 use facet_core::ScalarType;
1841 let scalar_shape = deref_pointer(inner_shape);
1842 match scalar_shape.scalar_type() {
1843 Some(ScalarType::Bool) => {
1844 result.bool_variants.push((variant, inner_shape));
1845 }
1846 Some(
1847 ScalarType::U8
1848 | ScalarType::U16
1849 | ScalarType::U32
1850 | ScalarType::U64
1851 | ScalarType::U128
1852 | ScalarType::USize
1853 | ScalarType::I8
1854 | ScalarType::I16
1855 | ScalarType::I32
1856 | ScalarType::I64
1857 | ScalarType::I128
1858 | ScalarType::ISize,
1859 ) => {
1860 result.int_variants.push((variant, inner_shape));
1861 }
1862 Some(ScalarType::F32 | ScalarType::F64) => {
1863 result.float_variants.push((variant, inner_shape));
1864 }
1865 #[cfg(feature = "alloc")]
1866 Some(ScalarType::String | ScalarType::CowStr) => {
1867 result.string_variants.push((variant, inner_shape));
1868 }
1869 Some(ScalarType::Str | ScalarType::Char) => {
1870 result.string_variants.push((variant, inner_shape));
1871 }
1872 _ => {
1873 // Other scalar types (Unit, SocketAddr, IpAddr, etc.) - leave in general scalar_variants only
1874 }
1875 }
1876 }
1877 VariantFormat::NewtypeStruct { .. } => {
1878 result.struct_variants.push(variant);
1879 }
1880 VariantFormat::NewtypeTuple { arity, .. } => {
1881 result.tuple_variants.push((variant, arity));
1882 }
1883 VariantFormat::NewtypeSequence { .. } => {
1884 // Sequences like Vec<T> are variable-length, so we use arity 0
1885 // to indicate "accepts any array" (not an exact match requirement)
1886 result.tuple_variants.push((variant, 0));
1887 }
1888 VariantFormat::NewtypeOther { .. } => {
1889 result.other_variants.push(variant);
1890 }
1891 VariantFormat::Tuple { arity } => {
1892 result.tuple_variants.push((variant, arity));
1893 }
1894 VariantFormat::Struct => {
1895 result.struct_variants.push(variant);
1896 }
1897 }
1898 }
1899
1900 Some(result)
1901 }
1902
1903 /// Get tuple variants with a specific arity.
1904 pub fn tuple_variants_with_arity(&self, arity: usize) -> Vec<&'static Variant> {
1905 self.tuple_variants
1906 .iter()
1907 .filter(|(_, a)| *a == arity)
1908 .map(|(v, _)| *v)
1909 .collect()
1910 }
1911
1912 /// Check if there are any scalar-expecting variants.
1913 pub const fn has_scalar_variants(&self) -> bool {
1914 !self.scalar_variants.is_empty()
1915 }
1916
1917 /// Check if there are any tuple-expecting variants.
1918 pub const fn has_tuple_variants(&self) -> bool {
1919 !self.tuple_variants.is_empty()
1920 }
1921
1922 /// Check if there are any struct-expecting variants.
1923 pub const fn has_struct_variants(&self) -> bool {
1924 !self.struct_variants.is_empty()
1925 }
1926}
1927
1928// ============================================================================
1929// Schema Builder
1930// ============================================================================
1931
1932/// How enum variants are represented in the serialized format.
1933#[derive(Debug, Clone, PartialEq, Eq, Default)]
1934pub enum EnumRepr {
1935 /// Variant fields are flattened to the same level as other fields.
1936 /// Also used for `#[facet(untagged)]` enums where there's no tag at all.
1937 /// Used by formats like TOML where all fields appear at one level.
1938 /// Example: `{"name": "...", "host": "...", "port": 8080}`
1939 #[default]
1940 Flattened,
1941
1942 /// Variant name is a key, variant content is nested under it.
1943 /// This is the default serde representation for enums.
1944 /// Example: `{"name": "...", "Tcp": {"host": "...", "port": 8080}}`
1945 ExternallyTagged,
1946
1947 /// Tag field is inside the content, alongside variant fields.
1948 /// Used with `#[facet(tag = "type")]`.
1949 /// Example: `{"type": "Tcp", "host": "...", "port": 8080}`
1950 InternallyTagged {
1951 /// The name of the tag field (e.g., "type")
1952 tag: &'static str,
1953 },
1954
1955 /// Tag and content are adjacent fields at the same level.
1956 /// Used with `#[facet(tag = "t", content = "c")]`.
1957 /// Example: `{"t": "Tcp", "c": {"host": "...", "port": 8080}}`
1958 AdjacentlyTagged {
1959 /// The name of the tag field (e.g., "t")
1960 tag: &'static str,
1961 /// The name of the content field (e.g., "c")
1962 content: &'static str,
1963 },
1964}
1965
1966impl EnumRepr {
1967 /// Detect the enum representation from a Shape's attributes.
1968 ///
1969 /// Returns:
1970 /// - `Flattened` if `#[facet(untagged)]`
1971 /// - `InternallyTagged` if `#[facet(tag = "...")]` without content
1972 /// - `AdjacentlyTagged` if both `#[facet(tag = "...", content = "...")]`
1973 /// - `ExternallyTagged` if no attributes (the default enum representation)
1974 pub const fn from_shape(shape: &'static Shape) -> Self {
1975 let tag = shape.get_tag_attr();
1976 let content = shape.get_content_attr();
1977 let untagged = shape.is_untagged();
1978
1979 match (tag, content, untagged) {
1980 // Untagged explicitly requested
1981 (_, _, true) => EnumRepr::Flattened,
1982 // Both tag and content specified → adjacently tagged
1983 (Some(t), Some(c), false) => EnumRepr::AdjacentlyTagged { tag: t, content: c },
1984 // Only tag specified → internally tagged
1985 (Some(t), None, false) => EnumRepr::InternallyTagged { tag: t },
1986 // No attributes → default to externally tagged (variant name as key)
1987 (None, None, false) => EnumRepr::ExternallyTagged,
1988 // Content without tag is invalid, treat as externally tagged
1989 (None, Some(_), false) => EnumRepr::ExternallyTagged,
1990 }
1991 }
1992}
1993
1994impl Schema {
1995 /// Build a schema for the given shape with flattened enum representation.
1996 ///
1997 /// Returns an error if the type definition contains conflicts, such as
1998 /// duplicate field names from parent and flattened structs.
1999 ///
2000 /// Note: This defaults to `Flattened` representation. For auto-detection
2001 /// based on `#[facet(tag = "...")]` attributes, use [`Schema::build_auto`].
2002 pub fn build(shape: &'static Shape) -> Result<Self, SchemaError> {
2003 Self::build_with_repr(shape, EnumRepr::Flattened)
2004 }
2005
2006 /// Build a schema with auto-detected enum representation based on each enum's attributes.
2007 ///
2008 /// This inspects each flattened enum's shape attributes to determine its representation:
2009 /// - `#[facet(untagged)]` → Flattened
2010 /// - `#[facet(tag = "type")]` → InternallyTagged
2011 /// - `#[facet(tag = "t", content = "c")]` → AdjacentlyTagged
2012 /// - No attributes → Flattened (for flatten solver behavior)
2013 ///
2014 /// For externally-tagged enums (variant name as key), use [`Schema::build_externally_tagged`].
2015 pub fn build_auto(shape: &'static Shape) -> Result<Self, SchemaError> {
2016 let builder = SchemaBuilder::new(shape, EnumRepr::Flattened).with_auto_detect();
2017 builder.into_schema()
2018 }
2019
2020 /// Build a schema for externally-tagged enum representation (e.g., JSON).
2021 ///
2022 /// In this representation, the variant name appears as a key and the
2023 /// variant's content is nested under it. The solver will only expect
2024 /// to see the variant name as a top-level key, not the variant's fields.
2025 pub fn build_externally_tagged(shape: &'static Shape) -> Result<Self, SchemaError> {
2026 Self::build_with_repr(shape, EnumRepr::ExternallyTagged)
2027 }
2028
2029 /// Build a schema with the specified enum representation.
2030 pub fn build_with_repr(shape: &'static Shape, repr: EnumRepr) -> Result<Self, SchemaError> {
2031 let builder = SchemaBuilder::new(shape, repr);
2032 builder.into_schema()
2033 }
2034
2035 /// Get the resolutions for this schema.
2036 pub fn resolutions(&self) -> &[Resolution] {
2037 &self.resolutions
2038 }
2039
2040 /// Get the format this schema was built for.
2041 pub const fn format(&self) -> Format {
2042 self.format
2043 }
2044
2045 /// Build a schema for DOM format (XML, HTML) with auto-detected enum representation.
2046 ///
2047 /// In DOM format, fields are categorized as attributes, elements, or text content.
2048 /// The solver uses `see_attribute()`, `see_element()`, etc. to report fields.
2049 pub fn build_dom(shape: &'static Shape) -> Result<Self, SchemaError> {
2050 let builder = SchemaBuilder::new(shape, EnumRepr::Flattened)
2051 .with_auto_detect()
2052 .with_format(Format::Dom);
2053 builder.into_schema()
2054 }
2055
2056 /// Build a schema with a specific format.
2057 pub fn build_with_format(shape: &'static Shape, format: Format) -> Result<Self, SchemaError> {
2058 let builder = SchemaBuilder::new(shape, EnumRepr::Flattened)
2059 .with_auto_detect()
2060 .with_format(format);
2061 builder.into_schema()
2062 }
2063}
2064
2065struct SchemaBuilder {
2066 shape: &'static Shape,
2067 enum_repr: EnumRepr,
2068 /// If true, detect enum representation from each enum's shape attributes.
2069 /// If false, use `enum_repr` for all enums.
2070 auto_detect_enum_repr: bool,
2071 /// The format to build the schema for.
2072 format: Format,
2073}
2074
2075impl SchemaBuilder {
2076 const fn new(shape: &'static Shape, enum_repr: EnumRepr) -> Self {
2077 Self {
2078 shape,
2079 enum_repr,
2080 auto_detect_enum_repr: false,
2081 format: Format::Flat,
2082 }
2083 }
2084
2085 const fn with_auto_detect(mut self) -> Self {
2086 self.auto_detect_enum_repr = true;
2087 self
2088 }
2089
2090 const fn with_format(mut self, format: Format) -> Self {
2091 self.format = format;
2092 self
2093 }
2094
2095 fn analyze(&self) -> Result<Vec<Resolution>, SchemaError> {
2096 self.analyze_shape(self.shape, FieldPath::empty(), Vec::new())
2097 }
2098
2099 /// Analyze a shape and return all possible resolutions.
2100 /// Returns a Vec because enums create multiple resolutions.
2101 ///
2102 /// - `current_path`: The internal field path (for FieldInfo)
2103 /// - `key_prefix`: The serialized key path prefix (for known_paths)
2104 fn analyze_shape(
2105 &self,
2106 shape: &'static Shape,
2107 current_path: FieldPath,
2108 key_prefix: KeyPath,
2109 ) -> Result<Vec<Resolution>, SchemaError> {
2110 match shape.ty {
2111 Type::User(UserType::Struct(struct_type)) => {
2112 self.analyze_struct(struct_type, current_path, key_prefix)
2113 }
2114 Type::User(UserType::Enum(enum_type)) => {
2115 // Enum at root level: create one configuration per variant
2116 self.analyze_enum(shape, enum_type, current_path, key_prefix)
2117 }
2118 _ => {
2119 // For non-struct types at root level, return single empty config
2120 Ok(vec![Resolution::new()])
2121 }
2122 }
2123 }
2124
2125 /// Analyze an enum and return one configuration per variant.
2126 ///
2127 /// - `current_path`: The internal field path (for FieldInfo)
2128 /// - `key_prefix`: The serialized key path prefix (for known_paths)
2129 fn analyze_enum(
2130 &self,
2131 shape: &'static Shape,
2132 enum_type: facet_core::EnumType,
2133 current_path: FieldPath,
2134 key_prefix: KeyPath,
2135 ) -> Result<Vec<Resolution>, SchemaError> {
2136 let enum_name = shape.type_identifier;
2137 let mut result = Vec::new();
2138
2139 for variant in enum_type.variants {
2140 let mut config = Resolution::new();
2141
2142 // Record this variant selection
2143 config.add_variant_selection(current_path.clone(), enum_name, variant.name);
2144
2145 let variant_path = current_path.push_variant("", variant.name);
2146
2147 // Get resolutions from the variant's content
2148 let variant_configs =
2149 self.analyze_variant_content(variant, &variant_path, &key_prefix)?;
2150
2151 // Merge each variant config into the base
2152 for variant_config in variant_configs {
2153 let mut final_config = config.clone();
2154 final_config.merge(&variant_config)?;
2155 result.push(final_config);
2156 }
2157 }
2158
2159 Ok(result)
2160 }
2161
2162 /// Analyze a struct and return all possible resolutions.
2163 ///
2164 /// - `current_path`: The internal field path (for FieldInfo)
2165 /// - `key_prefix`: The serialized key path prefix (for known_paths)
2166 fn analyze_struct(
2167 &self,
2168 struct_type: StructType,
2169 current_path: FieldPath,
2170 key_prefix: KeyPath,
2171 ) -> Result<Vec<Resolution>, SchemaError> {
2172 // Start with one empty configuration
2173 let mut configs = vec![Resolution::new()];
2174
2175 // Process each field, potentially multiplying resolutions
2176 for field in struct_type.fields {
2177 configs =
2178 self.analyze_field_into_configs(field, ¤t_path, &key_prefix, configs)?;
2179 }
2180
2181 Ok(configs)
2182 }
2183
2184 /// Process a field and return updated resolutions.
2185 /// If the field is a flattened enum, this may multiply the number of configs.
2186 ///
2187 /// - `parent_path`: The internal field path to the parent (for FieldInfo)
2188 /// - `key_prefix`: The serialized key path prefix (for known_paths)
2189 fn analyze_field_into_configs(
2190 &self,
2191 field: &'static Field,
2192 parent_path: &FieldPath,
2193 key_prefix: &KeyPath,
2194 mut configs: Vec<Resolution>,
2195 ) -> Result<Vec<Resolution>, SchemaError> {
2196 let is_flatten = field.is_flattened();
2197
2198 if is_flatten {
2199 // Flattened: inner keys bubble up to current level (same key_prefix)
2200 self.analyze_flattened_field_into_configs(field, parent_path, key_prefix, configs)
2201 } else {
2202 // Regular field: add to ALL current configs
2203 let field_path = parent_path.push_field(field.name);
2204 let required = !field.has_default() && !is_option_type(field.shape());
2205
2206 // Build the key path for this field (uses effective_name for wire format)
2207 let mut field_key_path = key_prefix.clone();
2208 field_key_path.push(field.effective_name());
2209
2210 let field_info = FieldInfo {
2211 serialized_name: field.effective_name(),
2212 path: field_path,
2213 required,
2214 value_shape: field.shape(),
2215 field,
2216 category: if self.format == Format::Dom {
2217 FieldCategory::from_field_dom(field).unwrap_or(FieldCategory::Element)
2218 } else {
2219 FieldCategory::Flat
2220 },
2221 };
2222
2223 for config in &mut configs {
2224 config.add_field(field_info.clone())?;
2225 // Add this field's key path
2226 config.add_key_path(field_key_path.clone());
2227 }
2228
2229 // If the field's value is a struct, recurse to collect nested key paths
2230 // (for probing, not for flattening - these are nested in serialized format)
2231 // This may fork resolutions if the nested struct contains flattened enums!
2232 configs =
2233 self.collect_nested_key_paths_for_shape(field.shape(), &field_key_path, configs)?;
2234
2235 Ok(configs)
2236 }
2237 }
2238
2239 /// Collect nested key paths from a shape into resolutions.
2240 /// This handles the case where a non-flattened field contains a struct with flattened enums.
2241 /// Returns updated resolutions (may fork if flattened enums are encountered).
2242 fn collect_nested_key_paths_for_shape(
2243 &self,
2244 shape: &'static Shape,
2245 key_prefix: &KeyPath,
2246 configs: Vec<Resolution>,
2247 ) -> Result<Vec<Resolution>, SchemaError> {
2248 match shape.ty {
2249 Type::User(UserType::Struct(struct_type)) => {
2250 self.collect_nested_key_paths_for_struct(struct_type, key_prefix, configs)
2251 }
2252 _ => Ok(configs),
2253 }
2254 }
2255
2256 /// Collect nested key paths from a struct, potentially forking for flattened enums.
2257 fn collect_nested_key_paths_for_struct(
2258 &self,
2259 struct_type: StructType,
2260 key_prefix: &KeyPath,
2261 mut configs: Vec<Resolution>,
2262 ) -> Result<Vec<Resolution>, SchemaError> {
2263 for field in struct_type.fields {
2264 let is_flatten = field.is_flattened();
2265 let mut field_key_path = key_prefix.clone();
2266
2267 if is_flatten {
2268 // Flattened field: keys bubble up to current level, may fork configs
2269 configs =
2270 self.collect_nested_key_paths_for_flattened(field, key_prefix, configs)?;
2271 } else {
2272 // Regular field: add key path and recurse
2273 field_key_path.push(field.effective_name());
2274
2275 for config in &mut configs {
2276 config.add_key_path(field_key_path.clone());
2277 }
2278
2279 // Recurse into nested structs
2280 configs = self.collect_nested_key_paths_for_shape(
2281 field.shape(),
2282 &field_key_path,
2283 configs,
2284 )?;
2285 }
2286 }
2287 Ok(configs)
2288 }
2289
2290 /// Handle flattened fields when collecting nested key paths.
2291 /// This may fork resolutions for flattened enums.
2292 fn collect_nested_key_paths_for_flattened(
2293 &self,
2294 field: &'static Field,
2295 key_prefix: &KeyPath,
2296 configs: Vec<Resolution>,
2297 ) -> Result<Vec<Resolution>, SchemaError> {
2298 let shape = field.shape();
2299
2300 match shape.ty {
2301 Type::User(UserType::Struct(struct_type)) => {
2302 // Flattened struct: recurse with same key_prefix
2303 self.collect_nested_key_paths_for_struct(struct_type, key_prefix, configs)
2304 }
2305 Type::User(UserType::Enum(enum_type)) => {
2306 // Flattened enum: fork resolutions
2307 // We need to match each config to its corresponding variant
2308 let mut result = Vec::new();
2309
2310 for config in configs {
2311 // Find which variant this config has selected for this field
2312 let selected_variant = config
2313 .variant_selections()
2314 .iter()
2315 .find(|vs| {
2316 // Match by the field name in the path
2317 vs.path.segments().last() == Some(&PathSegment::Field(field.name))
2318 })
2319 .map(|vs| vs.variant_name);
2320
2321 if let Some(variant_name) = selected_variant {
2322 // Find the variant and collect its key paths
2323 if let Some(variant) =
2324 enum_type.variants.iter().find(|v| v.name == variant_name)
2325 {
2326 let mut updated_config = config;
2327 updated_config = self.collect_variant_key_paths(
2328 variant,
2329 key_prefix,
2330 updated_config,
2331 )?;
2332 result.push(updated_config);
2333 } else {
2334 result.push(config);
2335 }
2336 } else {
2337 result.push(config);
2338 }
2339 }
2340 Ok(result)
2341 }
2342 _ => Ok(configs),
2343 }
2344 }
2345
2346 /// Collect key paths from an enum variant's content.
2347 fn collect_variant_key_paths(
2348 &self,
2349 variant: &'static Variant,
2350 key_prefix: &KeyPath,
2351 mut config: Resolution,
2352 ) -> Result<Resolution, SchemaError> {
2353 // Check if this is a newtype variant (single unnamed field)
2354 if variant.data.fields.len() == 1 && variant.data.fields[0].name == "0" {
2355 let inner_field = &variant.data.fields[0];
2356 let inner_shape = inner_field.shape();
2357
2358 // If the inner type is a struct, flatten its fields
2359 if let Type::User(UserType::Struct(inner_struct)) = inner_shape.ty {
2360 let configs = self.collect_nested_key_paths_for_struct(
2361 inner_struct,
2362 key_prefix,
2363 vec![config],
2364 )?;
2365 return Ok(configs.into_iter().next().unwrap_or_else(Resolution::new));
2366 }
2367 }
2368
2369 // Named fields - process each
2370 for variant_field in variant.data.fields {
2371 let is_flatten = variant_field.is_flattened();
2372
2373 if is_flatten {
2374 let configs = self.collect_nested_key_paths_for_flattened(
2375 variant_field,
2376 key_prefix,
2377 vec![config],
2378 )?;
2379 config = configs.into_iter().next().unwrap_or_else(Resolution::new);
2380 } else {
2381 let mut field_key_path = key_prefix.clone();
2382 field_key_path.push(variant_field.effective_name());
2383 config.add_key_path(field_key_path.clone());
2384
2385 let configs = self.collect_nested_key_paths_for_shape(
2386 variant_field.shape(),
2387 &field_key_path,
2388 vec![config],
2389 )?;
2390 config = configs.into_iter().next().unwrap_or_else(Resolution::new);
2391 }
2392 }
2393 Ok(config)
2394 }
2395
2396 /// Collect ONLY key paths from a variant's content (no fields added).
2397 /// Used for externally-tagged enums where variant content is nested and
2398 /// will be parsed separately by the deserializer.
2399 fn collect_variant_key_paths_only(
2400 &self,
2401 variant: &'static Variant,
2402 key_prefix: &KeyPath,
2403 config: &mut Resolution,
2404 ) -> Result<(), SchemaError> {
2405 Self::collect_variant_fields_key_paths_only(variant, key_prefix, config);
2406 Ok(())
2407 }
2408
2409 /// Recursively collect key paths from a struct (no fields added).
2410 fn collect_struct_key_paths_only(
2411 struct_type: StructType,
2412 key_prefix: &KeyPath,
2413 config: &mut Resolution,
2414 ) {
2415 for field in struct_type.fields {
2416 let is_flatten = field.is_flattened();
2417
2418 if is_flatten {
2419 // Flattened field: keys bubble up to current level
2420 Self::collect_shape_key_paths_only(field.shape(), key_prefix, config);
2421 } else {
2422 // Regular field: add its key path
2423 let mut field_key_path = key_prefix.clone();
2424 field_key_path.push(field.effective_name());
2425 config.add_key_path(field_key_path.clone());
2426
2427 // Recurse into nested types
2428 Self::collect_shape_key_paths_only(field.shape(), &field_key_path, config);
2429 }
2430 }
2431 }
2432
2433 /// Recursively collect key paths from a shape (struct or enum).
2434 fn collect_shape_key_paths_only(
2435 shape: &'static Shape,
2436 key_prefix: &KeyPath,
2437 config: &mut Resolution,
2438 ) {
2439 match shape.ty {
2440 Type::User(UserType::Struct(inner_struct)) => {
2441 Self::collect_struct_key_paths_only(inner_struct, key_prefix, config);
2442 }
2443 Type::User(UserType::Enum(enum_type)) => {
2444 // For enums, collect key paths from ALL variants
2445 // (we don't know which variant will be selected)
2446 for variant in enum_type.variants {
2447 Self::collect_variant_fields_key_paths_only(variant, key_prefix, config);
2448 }
2449 }
2450 _ => {}
2451 }
2452 }
2453
2454 /// Collect key paths from a variant's fields (not the variant itself).
2455 fn collect_variant_fields_key_paths_only(
2456 variant: &'static Variant,
2457 key_prefix: &KeyPath,
2458 config: &mut Resolution,
2459 ) {
2460 // Check if this is a newtype variant (single unnamed field)
2461 if variant.data.fields.len() == 1 && variant.data.fields[0].name == "0" {
2462 let inner_field = &variant.data.fields[0];
2463 Self::collect_shape_key_paths_only(inner_field.shape(), key_prefix, config);
2464 return;
2465 }
2466
2467 // Named fields - add key paths for each
2468 for variant_field in variant.data.fields {
2469 let mut field_key_path = key_prefix.clone();
2470 field_key_path.push(variant_field.effective_name());
2471 config.add_key_path(field_key_path.clone());
2472
2473 // Recurse into nested types
2474 Self::collect_shape_key_paths_only(variant_field.shape(), &field_key_path, config);
2475 }
2476 }
2477
2478 /// Process a flattened field, potentially forking resolutions for enums.
2479 ///
2480 /// For flattened fields, the inner keys bubble up to the current level,
2481 /// so we pass the same key_prefix (not key_prefix + field.name).
2482 ///
2483 /// If the field is `Option<T>`, we unwrap to get T and mark all resulting
2484 /// fields as optional (since the entire flattened block can be omitted).
2485 fn analyze_flattened_field_into_configs(
2486 &self,
2487 field: &'static Field,
2488 parent_path: &FieldPath,
2489 key_prefix: &KeyPath,
2490 configs: Vec<Resolution>,
2491 ) -> Result<Vec<Resolution>, SchemaError> {
2492 let field_path = parent_path.push_field(field.name);
2493 let original_shape = field.shape();
2494
2495 // Check if this is Option<T> - if so, unwrap and mark all fields optional
2496 let (shape, is_optional_flatten) = match unwrap_option_type(original_shape) {
2497 Some(inner) => (inner, true),
2498 None => (original_shape, false),
2499 };
2500
2501 match shape.ty {
2502 Type::User(UserType::Struct(struct_type)) => {
2503 // Flatten a struct: get its resolutions and merge into each of ours
2504 // Key prefix stays the same - inner keys bubble up
2505 let mut struct_configs =
2506 self.analyze_struct(struct_type, field_path, key_prefix.clone())?;
2507
2508 // If the flatten field was Option<T>, mark all inner fields as optional
2509 if is_optional_flatten {
2510 for config in &mut struct_configs {
2511 config.mark_all_optional();
2512 }
2513 }
2514
2515 // Each of our configs combines with each struct config
2516 // (usually struct_configs has 1 element unless it contains enums)
2517 let mut result = Vec::new();
2518 for base_config in configs {
2519 for struct_config in &struct_configs {
2520 let mut merged = base_config.clone();
2521 merged.merge(struct_config)?;
2522 result.push(merged);
2523 }
2524 }
2525 Ok(result)
2526 }
2527 Type::User(UserType::Enum(enum_type)) => {
2528 // Fork: each existing config × each variant
2529 let mut result = Vec::new();
2530 let enum_name = shape.type_identifier;
2531
2532 // Determine enum representation:
2533 // - If auto_detect_enum_repr is enabled, detect from the enum's shape attributes
2534 // - Otherwise, use the global enum_repr setting
2535 let enum_repr = if self.auto_detect_enum_repr {
2536 EnumRepr::from_shape(shape)
2537 } else {
2538 self.enum_repr.clone()
2539 };
2540
2541 for base_config in configs {
2542 for variant in enum_type.variants {
2543 let mut forked = base_config.clone();
2544 forked.add_variant_selection(field_path.clone(), enum_name, variant.name);
2545
2546 let variant_path = field_path.push_variant(field.name, variant.name);
2547
2548 match &enum_repr {
2549 EnumRepr::ExternallyTagged => {
2550 // For externally tagged enums, the variant name is a key
2551 // at the current level, and its content is nested underneath.
2552 let mut variant_key_prefix = key_prefix.clone();
2553 variant_key_prefix.push(variant.name);
2554
2555 // Add the variant name itself as a known key path
2556 forked.add_key_path(variant_key_prefix.clone());
2557
2558 // Add the variant name as a field (the key that selects this variant)
2559 let variant_field_info = FieldInfo {
2560 serialized_name: variant.name,
2561 path: variant_path.clone(),
2562 required: !is_optional_flatten,
2563 value_shape: shape, // The enum shape
2564 field, // The original flatten field
2565 category: FieldCategory::Element, // Variant selector is like an element
2566 };
2567 forked.add_field(variant_field_info)?;
2568
2569 // For externally-tagged enums, we do NOT add the variant's
2570 // inner fields to required fields. They're nested and will
2571 // be parsed separately by the deserializer.
2572 // Only add them to known_paths for depth-aware probing.
2573 self.collect_variant_key_paths_only(
2574 variant,
2575 &variant_key_prefix,
2576 &mut forked,
2577 )?;
2578
2579 result.push(forked);
2580 }
2581 EnumRepr::Flattened => {
2582 // For flattened/untagged enums, the variant's fields appear at the
2583 // same level as other fields. The variant name is NOT a key;
2584 // only the variant's inner fields are keys.
2585
2586 // Get resolutions from the variant's content
2587 // Key prefix stays the same - inner keys bubble up
2588 let mut variant_configs = self.analyze_variant_content(
2589 variant,
2590 &variant_path,
2591 key_prefix,
2592 )?;
2593
2594 // If the flatten field was Option<T>, mark all inner fields as optional
2595 if is_optional_flatten {
2596 for config in &mut variant_configs {
2597 config.mark_all_optional();
2598 }
2599 }
2600
2601 // Merge each variant config into the forked base
2602 for variant_config in variant_configs {
2603 let mut final_config = forked.clone();
2604 final_config.merge(&variant_config)?;
2605 result.push(final_config);
2606 }
2607 }
2608 EnumRepr::InternallyTagged { tag } => {
2609 // For internally tagged enums, the tag field appears at the
2610 // same level as the variant's fields.
2611 // Example: {"type": "Tcp", "host": "...", "port": 8080}
2612
2613 // Add the tag field as a known key path
2614 let mut tag_key_path = key_prefix.clone();
2615 tag_key_path.push(tag);
2616 forked.add_key_path(tag_key_path);
2617
2618 // Add the tag field info - the tag discriminates the variant
2619 // We use a synthetic field for the tag
2620 let tag_field_info = FieldInfo {
2621 serialized_name: tag,
2622 path: variant_path.clone(),
2623 required: !is_optional_flatten,
2624 value_shape: shape, // The enum shape
2625 field, // The original flatten field
2626 category: FieldCategory::Element, // Tag is a key field
2627 };
2628 forked.add_field(tag_field_info)?;
2629
2630 // Get resolutions from the variant's content
2631 // Key prefix stays the same - inner keys are at the same level
2632 let mut variant_configs = self.analyze_variant_content(
2633 variant,
2634 &variant_path,
2635 key_prefix,
2636 )?;
2637
2638 // If the flatten field was Option<T>, mark all inner fields as optional
2639 if is_optional_flatten {
2640 for config in &mut variant_configs {
2641 config.mark_all_optional();
2642 }
2643 }
2644
2645 // Merge each variant config into the forked base
2646 for variant_config in variant_configs {
2647 let mut final_config = forked.clone();
2648 final_config.merge(&variant_config)?;
2649 result.push(final_config);
2650 }
2651 }
2652 EnumRepr::AdjacentlyTagged { tag, content } => {
2653 // For adjacently tagged enums, both tag and content fields
2654 // appear at the same level. Content contains the variant's fields.
2655 // Example: {"t": "Tcp", "c": {"host": "...", "port": 8080}}
2656
2657 // Add the tag field as a known key path
2658 let mut tag_key_path = key_prefix.clone();
2659 tag_key_path.push(tag);
2660 forked.add_key_path(tag_key_path);
2661
2662 // Add the tag field info
2663 let tag_field_info = FieldInfo {
2664 serialized_name: tag,
2665 path: variant_path.clone(),
2666 required: !is_optional_flatten,
2667 value_shape: shape, // The enum shape
2668 field, // The original flatten field
2669 category: FieldCategory::Element, // Tag is a key field
2670 };
2671 forked.add_field(tag_field_info)?;
2672
2673 // Add the content field as a known key path
2674 let mut content_key_prefix = key_prefix.clone();
2675 content_key_prefix.push(content);
2676 forked.add_key_path(content_key_prefix.clone());
2677
2678 // The variant's fields are nested under the content key
2679 // Collect key paths for probing
2680 self.collect_variant_key_paths_only(
2681 variant,
2682 &content_key_prefix,
2683 &mut forked,
2684 )?;
2685
2686 result.push(forked);
2687 }
2688 }
2689 }
2690 }
2691 Ok(result)
2692 }
2693 _ => {
2694 // Check if this is a Map type - if so, it becomes a catch-all for unknown fields
2695 if let Def::Map(map_def) = &shape.def {
2696 // Validate: key must be String for catch-all maps
2697 if map_def.k.scalar_type() == Some(facet_core::ScalarType::String) {
2698 // This is a valid catch-all map
2699 let field_info = FieldInfo {
2700 serialized_name: field.effective_name(),
2701 path: field_path,
2702 required: false, // Catch-all maps are never required
2703 value_shape: shape,
2704 field,
2705 // For DOM format, determine if this catches attributes or elements
2706 // based on the field's attributes
2707 category: if self.format == Format::Dom {
2708 if field.is_attribute() {
2709 FieldCategory::Attribute
2710 } else {
2711 FieldCategory::Element
2712 }
2713 } else {
2714 FieldCategory::Flat
2715 },
2716 };
2717
2718 let mut result = configs;
2719 for config in &mut result {
2720 config.set_catch_all_map(field_info.category, field_info.clone());
2721 }
2722 return Ok(result);
2723 }
2724 }
2725
2726 // Check if this is a DynamicValue type (like facet_value::Value) - also a catch-all
2727 if matches!(&shape.def, Def::DynamicValue(_)) {
2728 let field_info = FieldInfo {
2729 serialized_name: field.effective_name(),
2730 path: field_path,
2731 required: false, // Catch-all dynamic values are never required
2732 value_shape: shape,
2733 field,
2734 category: if self.format == Format::Dom {
2735 if field.is_attribute() {
2736 FieldCategory::Attribute
2737 } else {
2738 FieldCategory::Element
2739 }
2740 } else {
2741 FieldCategory::Flat
2742 },
2743 };
2744
2745 let mut result = configs;
2746 for config in &mut result {
2747 config.set_catch_all_map(field_info.category, field_info.clone());
2748 }
2749 return Ok(result);
2750 }
2751
2752 // Can't flatten other types - treat as regular field
2753 // For Option<T> flatten, also consider optionality from the wrapper
2754 let required =
2755 !field.has_default() && !is_option_type(shape) && !is_optional_flatten;
2756
2757 // For non-flattenable types, add the field with its key path
2758 let mut field_key_path = key_prefix.clone();
2759 field_key_path.push(field.effective_name());
2760
2761 let field_info = FieldInfo {
2762 serialized_name: field.effective_name(),
2763 path: field_path,
2764 required,
2765 value_shape: shape,
2766 field,
2767 category: if self.format == Format::Dom {
2768 FieldCategory::from_field_dom(field).unwrap_or(FieldCategory::Element)
2769 } else {
2770 FieldCategory::Flat
2771 },
2772 };
2773
2774 let mut result = configs;
2775 for config in &mut result {
2776 config.add_field(field_info.clone())?;
2777 config.add_key_path(field_key_path.clone());
2778 }
2779 Ok(result)
2780 }
2781 }
2782 }
2783
2784 /// Analyze a variant's content and return resolutions.
2785 ///
2786 /// - `variant_path`: The internal field path (for FieldInfo)
2787 /// - `key_prefix`: The serialized key path prefix (for known_paths)
2788 fn analyze_variant_content(
2789 &self,
2790 variant: &'static Variant,
2791 variant_path: &FieldPath,
2792 key_prefix: &KeyPath,
2793 ) -> Result<Vec<Resolution>, SchemaError> {
2794 // Check if this is a newtype variant (single unnamed field like `Foo(Bar)`)
2795 if variant.data.fields.len() == 1 && variant.data.fields[0].name == "0" {
2796 let inner_field = &variant.data.fields[0];
2797 let inner_shape = inner_field.shape();
2798
2799 // If the inner type is a struct, treat the newtype wrapper as transparent.
2800 //
2801 // Previously we pushed a synthetic `"0"` segment onto the path. That made the
2802 // solver think there was an extra field between the variant and the inner
2803 // struct (e.g., `backend.backend::Local.0.cache`). Format-specific flattening does not
2804 // expose that tuple wrapper, so the deserializer would try to open a field
2805 // named `"0"` on the inner struct/enum, causing "no such field" errors when
2806 // navigating paths like `backend::Local.cache`.
2807 //
2808 // Keep the synthetic `"0"` segment so the solver/reflect layer walks through
2809 // the tuple wrapper that Rust generates for newtype variants.
2810
2811 // For untagged enum variant resolution, we need to look at the "effective"
2812 // shape that determines the serialization format. This unwraps:
2813 // 1. Transparent wrappers (shape.inner) - e.g., `Curve64(GCurve<f64, f64>)`
2814 // 2. Proxy types (shape.proxy) - e.g., `GCurve` uses `GCurveProxy` for ser/de
2815 //
2816 // This ensures that `{"x":..., "y":...}` correctly matches `Linear(Curve64)`
2817 // where Curve64 is transparent around GCurve which has a proxy with x,y fields.
2818 let effective_shape = unwrap_to_effective_shape(inner_shape);
2819
2820 if let Type::User(UserType::Struct(inner_struct)) = effective_shape.ty {
2821 let inner_path = variant_path.push_field("0");
2822 return self.analyze_struct(inner_struct, inner_path, key_prefix.clone());
2823 }
2824 }
2825
2826 // Named fields or multiple fields - analyze as a pseudo-struct
2827 let mut configs = vec![Resolution::new()];
2828 for variant_field in variant.data.fields {
2829 configs =
2830 self.analyze_field_into_configs(variant_field, variant_path, key_prefix, configs)?;
2831 }
2832 Ok(configs)
2833 }
2834
2835 fn into_schema(self) -> Result<Schema, SchemaError> {
2836 let resolutions = self.analyze()?;
2837 let num_resolutions = resolutions.len();
2838
2839 // Build inverted index: field_name → bitmask of config indices (for Flat format)
2840 let mut field_to_resolutions: BTreeMap<&'static str, ResolutionSet> = BTreeMap::new();
2841 for (idx, config) in resolutions.iter().enumerate() {
2842 for field_info in config.fields().values() {
2843 field_to_resolutions
2844 .entry(field_info.serialized_name)
2845 .or_insert_with(|| ResolutionSet::empty(num_resolutions))
2846 .insert(idx);
2847 }
2848 }
2849
2850 // Build DOM inverted index: (category, name) → bitmask of config indices
2851 let mut dom_field_to_resolutions: BTreeMap<(FieldCategory, &'static str), ResolutionSet> =
2852 BTreeMap::new();
2853 if self.format == Format::Dom {
2854 for (idx, config) in resolutions.iter().enumerate() {
2855 for field_info in config.fields().values() {
2856 dom_field_to_resolutions
2857 .entry((field_info.category, field_info.serialized_name))
2858 .or_insert_with(|| ResolutionSet::empty(num_resolutions))
2859 .insert(idx);
2860 }
2861 }
2862 }
2863
2864 Ok(Schema {
2865 shape: self.shape,
2866 format: self.format,
2867 resolutions,
2868 field_to_resolutions,
2869 dom_field_to_resolutions,
2870 })
2871 }
2872}
2873
2874/// Check if a shape represents an Option type.
2875const fn is_option_type(shape: &'static Shape) -> bool {
2876 matches!(shape.def, Def::Option(_))
2877}
2878
2879/// If shape is `Option<T>`, returns `Some(T's shape)`. Otherwise returns `None`.
2880const fn unwrap_option_type(shape: &'static Shape) -> Option<&'static Shape> {
2881 match shape.def {
2882 Def::Option(option_def) => Some(option_def.t),
2883 _ => None,
2884 }
2885}
2886
2887/// Unwrap transparent wrappers and proxies to get the effective shape for field matching.
2888///
2889/// When determining which untagged enum variant matches a set of fields, we need to
2890/// look at the "effective" shape that determines the serialization format:
2891///
2892/// 1. Transparent wrappers (shape.inner): e.g., `Curve64` wraps `GCurve<f64, f64>`
2893/// - The wrapper has no serialization presence; it serializes as its inner type
2894///
2895/// 2. Proxy types (shape.proxy): e.g., `GCurve` uses `GCurveProxy` for ser/de
2896/// - The proxy's fields are what appear in the serialized format
2897///
2898/// This function recursively unwraps these layers to find the shape whose fields
2899/// should be used for variant matching. For example:
2900/// - `Curve64` (transparent) → `GCurve<f64, f64>` (has proxy) → `GCurveProxy<f64, f64>`
2901fn unwrap_to_effective_shape(shape: &'static Shape) -> &'static Shape {
2902 // First, unwrap transparent wrappers
2903 let shape = unwrap_transparent(shape);
2904
2905 // Then, if there's a proxy, use its shape instead
2906 if let Some(proxy_def) = shape.proxy {
2907 // Recursively unwrap in case the proxy is also transparent or has its own proxy
2908 unwrap_to_effective_shape(proxy_def.shape)
2909 } else {
2910 shape
2911 }
2912}
2913
2914/// Recursively unwrap transparent wrappers to get to the innermost type.
2915fn unwrap_transparent(shape: &'static Shape) -> &'static Shape {
2916 if let Some(inner) = shape.inner {
2917 unwrap_transparent(inner)
2918 } else {
2919 shape
2920 }
2921}