@rip-lang/db 0.10.0 → 1.0.2

package/lib/duckdb-binary.rip

@@ -2,8 +2,6 @@
 #
 # Implements the binary serialization format used by DuckDB's official UI.
 # This allows rip-db to serve responses that the DuckDB UI can understand.
-#
-# Protocol spec: See PROTOCOL.md in this directory
 
 # ==============================================================================
 # LogicalTypeId - matches DuckDB's internal type IDs
@@ -46,6 +44,9 @@ export LogicalTypeId =
   UNION:        107
   ARRAY:        108
 
+# Shared TextEncoder instance (avoid allocating per call)
+textEncoder = new TextEncoder()
+
 # ==============================================================================
 # BinarySerializer - writes the DuckDB binary format
 # ==============================================================================
@@ -106,8 +107,7 @@ export class BinarySerializer
         break
 
   writeString: (str) ->
-    encoder = new TextEncoder()
-    bytes = encoder.encode str
+    bytes = textEncoder.encode str
     @writeVarInt bytes.length
     @buffer.push ...bytes
 
@@ -213,21 +213,23 @@ serializeType = (s, column) ->
 serializeDataChunk = (s, columns, rows) ->
   s.writePropertyVarInt 100, rows.length
   s.writeList 101, columns, (s, col, colIdx) ->
-    values = rows.map (row) -> row[colIdx] ? row[col.name]
+    values = rows.map (row) -> row[colIdx]
     serializeVector s, col, values
   s.writeEndMarker()
 
 serializeVector = (s, column, values) ->
   typeId = mapDuckDBType column.type
   hasNulls = values.some (v) -> v is null or v is undefined
-  allValid = if hasNulls then 0 else 1
 
+  # allValid flag: 0 = all valid (no bitmap), non-zero = has validity bitmap
+  # The deserializer reads field 101 ONLY when allValid is truthy
   s.writeFieldId 100
-  s.writeUint8 allValid
-
-  if not allValid
+  if hasNulls
+    s.writeUint8 1  # Has validity bitmap
     s.writeFieldId 101
     s.writeData createValidityBitmap values
+  else
+    s.writeUint8 0  # All valid, no bitmap needed
 
   switch typeId
     when LogicalTypeId.VARCHAR, LogicalTypeId.CHAR
@@ -329,6 +331,16 @@ serializeVector = (s, column, values) ->
         dv.setBigInt64 i * 8, micros, true
       s.writeData bytes
 
+    when LogicalTypeId.UUID
+      s.writeFieldId 102
+      bytes = new Uint8Array values.length * 16
+      dv = new DataView bytes.buffer
+      for v, i in values
+        { lo, hi } = uuidToHugeint(v)
+        dv.setBigUint64 i * 16, lo, true
+        dv.setBigInt64 i * 16 + 8, hi, true
+      s.writeData bytes
+
     else
       s.writeList 102, values, (s, v) -> s.writeString String(v ? '')
 
@@ -339,10 +351,11 @@ serializeVector = (s, column, values) ->
 # ==============================================================================
 
 createValidityBitmap = (values) ->
-  byteCount = Math.ceil values.length / 8
+  # Must be uint64-aligned (8-byte chunks) — the UI reads validity with getBigUint64
+  byteCount = Math.ceil(values.length / 64) * 8
   bitmap = new Uint8Array byteCount
   for v, i in values
-    if v? and v isnt null
+    if v?
       byteIdx = Math.floor i / 8
       bitIdx = i % 8
       bitmap[byteIdx] |= 1 << bitIdx
@@ -370,6 +383,14 @@ timestampToMicros = (value) ->
   else
     0n
 
+uuidToHugeint = (uuid) ->
+  return { lo: 0n, hi: 0n } unless uuid
+  hex = String(uuid).replace(/-/g, '')
+  full = BigInt "0x#{hex}"
+  hi = (full >> 64n) ^ (1n << 63n)  # XOR sign bit for DuckDB sorting
+  lo = full & ((1n << 64n) - 1n)
+  { lo, hi }
+
 mapDuckDBType = (typeName) ->
   return LogicalTypeId.VARCHAR unless typeName
   upper = String(typeName).toUpperCase()
@@ -398,7 +419,7 @@ mapDuckDBType = (typeName) ->
     when 'INTERVAL' then LogicalTypeId.INTERVAL
     when 'JSON' then LogicalTypeId.VARCHAR
     else
-      if upper.startsWith 'DECIMAL' then LogicalTypeId.DOUBLE
+      if upper.startsWith 'DECIMAL' then LogicalTypeId.VARCHAR  # Serialize as string to preserve exact precision
       else if upper.startsWith 'VARCHAR' then LogicalTypeId.VARCHAR
       else if upper.startsWith 'CHAR' then LogicalTypeId.CHAR
       else LogicalTypeId.VARCHAR
@@ -520,6 +541,6 @@ export inferType = (value) ->
       else if /^\d{4}-\d{2}-\d{2}[T ]\d{2}:\d{2}/.test value then 'TIMESTAMP'
       else 'VARCHAR'
     when 'object'
-      if value instanceof Date then 'TIMESTAMP'
+      if value instanceof Date then 'TIMESTAMPTZ'
       else 'VARCHAR'
     else 'VARCHAR'