mongreldb_kit/
schema.rs

1//! Schema/value conversion between the kit model and MongrelDB core.
2
3use crate::error::{KitError, Result};
4use mongreldb_core::memtable::Value as CoreValue;
5use mongreldb_core::schema::{
6    ColumnDef, ColumnFlags, IndexDef, IndexKind, Schema as CoreSchema, TypeId,
7};
8use mongreldb_kit_core::schema::{
9    Column, ColumnType, IndexKind as KitIndexKind, Table as KitTable,
10};
11use serde_json::{Map, Value};
12
13/// Convert a kit table to a core schema.
14pub fn to_core_schema(table: &KitTable) -> CoreSchema {
15    let columns: Vec<ColumnDef> = table
16        .columns
17        .iter()
18        .map(|c| ColumnDef {
19            id: c.id as u16,
20            name: c.name.clone(),
21            ty: match c.storage_type {
22                ColumnType::Embedding => TypeId::Embedding {
23                    dim: c.embedding_dim.unwrap_or(0),
24                },
25                other => to_core_type(other),
26            },
27            flags: to_core_flags(table, c),
28        })
29        .collect();
30
31    let mut indexes: Vec<IndexDef> = Vec::new();
32    for idx in &table.indexes {
33        let kind = match idx.kind {
34            KitIndexKind::Bitmap => IndexKind::Bitmap,
35            KitIndexKind::Fm => IndexKind::FmIndex,
36            KitIndexKind::Ann => IndexKind::Ann,
37            KitIndexKind::Sparse => IndexKind::Sparse,
38            KitIndexKind::MinHash => IndexKind::MinHash,
39            KitIndexKind::LearnedRange => IndexKind::LearnedRange,
40        };
41        for col_name in &idx.columns {
42            if let Some(col) = table.column(col_name) {
43                indexes.push(IndexDef {
44                    name: format!("{}_{}", idx.name, col_name),
45                    column_id: col.id as u16,
46                    kind,
47                    predicate: None,
48                });
49            }
50        }
51    }
52    for uq in &table.unique_constraints {
53        for col_name in &uq.columns {
54            if let Some(col) = table.column(col_name) {
55                indexes.push(IndexDef {
56                    name: format!("uq_{}_{}", uq.name, col_name),
57                    column_id: col.id as u16,
58                    kind: IndexKind::Bitmap,
59                    predicate: None,
60                });
61            }
62        }
63    }
64
65    CoreSchema {
66        schema_id: table.id as u64,
67        columns,
68        indexes,
69        colocation: Vec::new(),
70        constraints: Default::default(),
71        clustered: false,
72    }
73}
74
75pub(crate) fn to_core_flags(table: &KitTable, column: &Column) -> ColumnFlags {
76    let mut flags = ColumnFlags::empty();
77    if column.nullable {
78        flags = flags.with(ColumnFlags::NULLABLE);
79    }
80    if table.primary_key.contains(&column.name) || column.primary_key {
81        flags = flags.with(ColumnFlags::PRIMARY_KEY);
82    }
83    if column.encrypted {
84        flags = flags.with(ColumnFlags::ENCRYPTED);
85    }
86    if column.encrypted_indexable {
87        flags = flags.with(ColumnFlags::ENCRYPTED_INDEXABLE);
88    }
89    flags
90}
91
92pub(crate) fn to_core_type(ty: ColumnType) -> TypeId {
93    match ty {
94        ColumnType::Bool => TypeId::Bool,
95        ColumnType::Int8 | ColumnType::Int16 | ColumnType::Int32 | ColumnType::Int64 => {
96            TypeId::Int64
97        }
98        ColumnType::Float32 | ColumnType::Float64 => TypeId::Float64,
99        ColumnType::Text
100        | ColumnType::Bytes
101        | ColumnType::Json
102        | ColumnType::Date
103        | ColumnType::DateTime => TypeId::Bytes,
104        ColumnType::TimestampNanos => TypeId::Int64,
105        ColumnType::Date64 => TypeId::Date64,
106        ColumnType::Time64 => TypeId::Time64,
107        ColumnType::Interval => TypeId::Interval,
108        ColumnType::Decimal128 => TypeId::Decimal128 {
109            precision: 38,
110            scale: 2,
111        },
112        ColumnType::Uuid => TypeId::Uuid,
113        ColumnType::JsonNative => TypeId::Json,
114        ColumnType::Array => TypeId::Array { element_type: 0 },
115        // Dimension is filled from the column's `embedding_dim` in
116        // `to_core_schema`; a bare type has no dimension context.
117        ColumnType::Embedding => TypeId::Embedding { dim: 0 },
118        // Sparse vectors are stored as bincode'd `Vec<(u32, f32)>` in a Bytes
119        // column; the Sparse index reads the tokens from those bytes.
120        ColumnType::Sparse => TypeId::Bytes,
121    }
122}
123
124/// Convert a JSON value to a core cell value using the column type for guidance.
125pub fn json_to_core(value: &Value, ty: ColumnType) -> Result<CoreValue> {
126    Ok(match value {
127        Value::Null => CoreValue::Null,
128        Value::Bool(b) => CoreValue::Bool(*b),
129        Value::Number(n) => {
130            if let Some(i) = n.as_i64() {
131                CoreValue::Int64(i)
132            } else {
133                CoreValue::Float64(n.as_f64().unwrap_or(f64::NAN))
134            }
135        }
136        Value::String(s) => CoreValue::Bytes(s.as_bytes().to_vec()),
137        Value::Array(arr) => {
138            if ty == ColumnType::Sparse {
139                let mut terms: Vec<(u32, f32)> = Vec::with_capacity(arr.len());
140                for pair in arr {
141                    let p = pair
142                        .as_array()
143                        .ok_or_else(|| KitError::Validation("sparse expects pairs".into()))?;
144                    let token =
145                        p.first().and_then(|v| v.as_u64()).ok_or_else(|| {
146                            KitError::Validation("sparse token must be u32".into())
147                        })? as u32;
148                    let weight = p.get(1).and_then(|v| v.as_f64()).ok_or_else(|| {
149                        KitError::Validation("sparse weight must be number".into())
150                    })? as f32;
151                    terms.push((token, weight));
152                }
153                CoreValue::Bytes(
154                    bincode::serialize(&terms).map_err(|e| KitError::Validation(e.to_string()))?,
155                )
156            } else if ty == ColumnType::Embedding {
157                let mut vec = Vec::with_capacity(arr.len());
158                for v in arr {
159                    match v.as_f64() {
160                        Some(f) => vec.push(f as f32),
161                        None => {
162                            return Err(KitError::Validation("embedding expects numbers".into()))
163                        }
164                    }
165                }
166                CoreValue::Embedding(vec)
167            } else if ty == ColumnType::Bytes {
168                let mut bytes = Vec::with_capacity(arr.len());
169                for v in arr {
170                    match v {
171                        Value::Number(n) => bytes.push(n.as_i64().unwrap_or(0) as u8),
172                        _ => return Err(KitError::Validation("bytes array expected".into())),
173                    }
174                }
175                CoreValue::Bytes(bytes)
176            } else {
177                CoreValue::Bytes(serde_json::to_vec(value)?)
178            }
179        }
180        Value::Object(_) => CoreValue::Bytes(serde_json::to_vec(value)?),
181    })
182}
183
184/// Convert a core cell value back to JSON, guided by the column type.
185pub fn core_to_json(value: &CoreValue, ty: ColumnType) -> Result<Value> {
186    Ok(match (value, ty) {
187        (CoreValue::Null, _) => Value::Null,
188        (CoreValue::Bool(b), _) => Value::Bool(*b),
189        (CoreValue::Int64(i), ColumnType::Int8) => Value::Number((*i as i8).into()),
190        (CoreValue::Int64(i), ColumnType::Int16) => Value::Number((*i as i16).into()),
191        (CoreValue::Int64(i), ColumnType::Int32) => Value::Number((*i as i32).into()),
192        (CoreValue::Int64(i), ColumnType::Int64) => Value::Number((*i).into()),
193        (CoreValue::Int64(i), ColumnType::TimestampNanos) => Value::Number((*i).into()),
194        (CoreValue::Int64(i), _) => Value::Number((*i).into()),
195        (CoreValue::Float64(f), ColumnType::Float32) => serde_json::to_value(*f as f32)?,
196        (CoreValue::Float64(f), _) => serde_json::to_value(*f)?,
197        (CoreValue::Bytes(b), ColumnType::Sparse) => {
198            let terms: Vec<(u32, f32)> =
199                bincode::deserialize(b).map_err(|e| KitError::Validation(e.to_string()))?;
200            Value::Array(
201                terms
202                    .into_iter()
203                    .map(|(t, w)| Value::Array(vec![Value::from(t), Value::from(w as f64)]))
204                    .collect(),
205            )
206        }
207        (CoreValue::Bytes(b), ColumnType::Bytes) => {
208            Value::Array(b.iter().map(|x| Value::Number((*x).into())).collect())
209        }
210        (CoreValue::Bytes(b), _) => match std::str::from_utf8(b) {
211            Ok(s) => Value::String(s.to_string()),
212            Err(_) => Value::Array(b.iter().map(|x| Value::Number((*x).into())).collect()),
213        },
214        (CoreValue::Embedding(v), _) => serde_json::to_value(v)?,
215        (CoreValue::Decimal(d), _) => Value::String(d.to_string()),
216        (
217            CoreValue::Interval {
218                months,
219                days,
220                nanos,
221            },
222            _,
223        ) => {
224            serde_json::json!({ "months": months, "days": days, "nanos": nanos })
225        }
226        (CoreValue::Uuid(b), _) => {
227            let hex: String = b.iter().map(|x| format!("{x:02x}")).collect();
228            serde_json::Value::String(hex)
229        }
230        (CoreValue::Json(b), _) => serde_json::from_slice(b.as_slice())
231            .unwrap_or_else(|_| serde_json::Value::String(String::from_utf8_lossy(b).into_owned())),
232    })
233}
234
235/// Build a JSON row from a core row and a kit table definition.
236pub fn core_row_to_json(row: &mongreldb_core::memtable::Row, table: &KitTable) -> Result<Row> {
237    let mut values = Map::new();
238    for col in &table.columns {
239        let v = row
240            .columns
241            .get(&(col.id as u16))
242            .cloned()
243            .unwrap_or(CoreValue::Null);
244        values.insert(col.name.clone(), core_to_json(&v, col.storage_type)?);
245    }
246    Ok(Row {
247        row_id: row.row_id.0,
248        values,
249    })
250}
251
252/// A kit row, identified by its internal storage row id and column values.
253#[derive(Debug, Clone, PartialEq)]
254pub struct Row {
255    pub row_id: u64,
256    pub values: Map<String, Value>,
257}
258
259impl Row {
260    /// Extract the primary-key value(s) as a JSON value.
261    ///
262    /// Single-column primary keys return the scalar value; composite keys return
263    /// an object.
264    pub fn pk(&self, table: &KitTable) -> Option<Value> {
265        if table.primary_key.len() == 1 {
266            self.values.get(&table.primary_key[0]).cloned()
267        } else {
268            let mut obj = Map::new();
269            for name in &table.primary_key {
270                obj.insert(
271                    name.clone(),
272                    self.values.get(name).cloned().unwrap_or(Value::Null),
273                );
274            }
275            Some(Value::Object(obj))
276        }
277    }
278}
279
280/// Extract the primary-key value(s) from a JSON value map.
281pub fn pk_value(values: &Map<String, Value>, table: &KitTable) -> Option<Value> {
282    if table.primary_key.len() == 1 {
283        values.get(&table.primary_key[0]).cloned()
284    } else {
285        let mut obj = Map::new();
286        for name in &table.primary_key {
287            obj.insert(
288                name.clone(),
289                values.get(name).cloned().unwrap_or(Value::Null),
290            );
291        }
292        Some(Value::Object(obj))
293    }
294}
295
296/// Convert a primary-key value into the column values for lookup.
297pub fn pk_to_map(pk: &Value, table: &KitTable) -> Result<Map<String, Value>> {
298    let mut map = Map::new();
299    match pk {
300        Value::Object(obj) => {
301            for name in &table.primary_key {
302                let v = obj
303                    .get(name)
304                    .cloned()
305                    .ok_or_else(|| KitError::Validation(format!("missing pk column {name}")))?;
306                map.insert(name.clone(), v);
307            }
308        }
309        scalar if table.primary_key.len() == 1 => {
310            map.insert(table.primary_key[0].clone(), scalar.clone());
311        }
312        _ => {
313            return Err(KitError::Validation(
314                "primary key value shape mismatch".into(),
315            ))
316        }
317    }
318    Ok(map)
319}
320
321/// Build a core cell vector from a JSON row and kit table definition.
322pub fn row_to_core_cells(
323    values: &Map<String, Value>,
324    table: &KitTable,
325) -> Result<Vec<(u16, CoreValue)>> {
326    let mut cells = Vec::with_capacity(table.columns.len());
327    for col in &table.columns {
328        let v = values.get(&col.name).cloned().unwrap_or(Value::Null);
329        cells.push((col.id as u16, json_to_core(&v, col.storage_type)?));
330    }
331    Ok(cells)
332}
mongreldb_kit/schema.rs

mongreldb_kit/
schema.rs
