oxedyne/fe2o3/fe2o3_text/tests/annealer_corpus/arrow_array.rs
48.8 KiB, 1 run
created by r1870400018:11686, which is this file's identity for as long as the history lasts, whatever it is later renamed to
download · who wrote it · its history
| 1 | // Licensed to the Apache Software Foundation (ASF) under one |
| 2 | // or more contributor license agreements. See the NOTICE file |
| 3 | // distributed with this work for additional information |
| 4 | // regarding copyright ownership. The ASF licenses this file |
| 5 | // to you under the Apache License, Version 2.0 (the |
| 6 | // "License"); you may not use this file except in compliance |
| 7 | // with the License. You may obtain a copy of the License at |
| 8 | // |
| 9 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | // |
| 11 | // Unless required by applicable law or agreed to in writing, |
| 12 | // software distributed under the License is distributed on an |
| 13 | // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
| 14 | // KIND, either express or implied. See the License for the |
| 15 | // specific language governing permissions and limitations |
| 16 | // under the License. |
| 17 | |
| 18 | //! The concrete array definitions |
| 19 | |
| 20 | mod binary_array; |
| 21 | |
| 22 | use crate::types::*; |
| 23 | use arrow_buffer::{ArrowNativeType, Buffer, NullBuffer, OffsetBuffer, ScalarBuffer}; |
| 24 | use arrow_data::ArrayData; |
| 25 | use arrow_schema::{DataType, IntervalUnit, TimeUnit}; |
| 26 | use std::any::Any; |
| 27 | use std::sync::Arc; |
| 28 | |
| 29 | pub use binary_array::*; |
| 30 | |
| 31 | mod boolean_array; |
| 32 | pub use boolean_array::*; |
| 33 | |
| 34 | mod byte_array; |
| 35 | pub use byte_array::*; |
| 36 | |
| 37 | mod dictionary_array; |
| 38 | pub use dictionary_array::*; |
| 39 | |
| 40 | mod fixed_size_binary_array; |
| 41 | pub use fixed_size_binary_array::*; |
| 42 | |
| 43 | mod fixed_size_list_array; |
| 44 | pub use fixed_size_list_array::*; |
| 45 | |
| 46 | mod list_array; |
| 47 | pub use list_array::*; |
| 48 | |
| 49 | mod map_array; |
| 50 | pub use map_array::*; |
| 51 | |
| 52 | mod null_array; |
| 53 | pub use null_array::*; |
| 54 | |
| 55 | mod primitive_array; |
| 56 | pub use primitive_array::*; |
| 57 | |
| 58 | mod string_array; |
| 59 | pub use string_array::*; |
| 60 | |
| 61 | mod struct_array; |
| 62 | pub use struct_array::*; |
| 63 | |
| 64 | mod union_array; |
| 65 | pub use union_array::*; |
| 66 | |
| 67 | mod run_array; |
| 68 | |
| 69 | pub use run_array::*; |
| 70 | |
| 71 | mod byte_view_array; |
| 72 | |
| 73 | pub use byte_view_array::*; |
| 74 | |
| 75 | mod list_view_array; |
| 76 | |
| 77 | pub use list_view_array::*; |
| 78 | |
| 79 | use crate::iterator::ArrayIter; |
| 80 | |
| 81 | /// An array in the [Arrow Columnar Format](https://arrow.apache.org/docs/format/Columnar.html) |
| 82 | /// |
| 83 | /// # Safety |
| 84 | /// |
| 85 | /// Implementations of this trait must ensure that all methods implementations comply with |
| 86 | /// the Arrow specification. No safety guards are placed and failing to comply with it can |
| 87 | /// translate into panics or undefined behavior. For example, a value computed based on `len` |
| 88 | /// may be used as a direct index into memory regions without checks. |
| 89 | /// |
| 90 | /// Note that it is likely impossible to correctly implement the trait for a |
| 91 | /// third party type, as substantial arrow-rs functionality is based on the |
| 92 | /// return values of [`Array::data_type`] and third party types cannot extend |
| 93 | /// the [`DataType`] enum. So any code that attempts casting based on data type |
| 94 | /// (including internal arrow library code) risks a panic or undefined behavior. |
| 95 | /// See [this discussion] for more details. |
| 96 | /// |
| 97 | /// This trait might be sealed in the future. Use at your own risk. |
| 98 | /// |
| 99 | /// [this discussion]: https://github.com/apache/arrow-rs/pull/9234#pullrequestreview-3708950936 |
| 100 | pub unsafe trait Array: std::fmt::Debug + Send + Sync { |
| 101 | /// Returns the array as [`Any`] so that it can be |
| 102 | /// downcasted to a specific implementation. |
| 103 | /// |
| 104 | /// # Example: |
| 105 | /// |
| 106 | /// ``` |
| 107 | /// # use std::sync::Arc; |
| 108 | /// # use arrow_array::{Int32Array, RecordBatch}; |
| 109 | /// # use arrow_schema::{Schema, Field, DataType, ArrowError}; |
| 110 | /// |
| 111 | /// let id = Int32Array::from(vec![1, 2, 3, 4, 5]); |
| 112 | /// let batch = RecordBatch::try_new( |
| 113 | /// Arc::new(Schema::new(vec![Field::new("id", DataType::Int32, false)])), |
| 114 | /// vec![Arc::new(id)] |
| 115 | /// ).unwrap(); |
| 116 | /// |
| 117 | /// let int32array = batch |
| 118 | /// .column(0) |
| 119 | /// .as_any() |
| 120 | /// .downcast_ref::<Int32Array>() |
| 121 | /// .expect("Failed to downcast"); |
| 122 | /// ``` |
| 123 | fn as_any(&self) -> &dyn Any; |
| 124 | |
| 125 | /// Returns the underlying data of this array |
| 126 | fn to_data(&self) -> ArrayData; |
| 127 | |
| 128 | /// Returns the underlying data of this array |
| 129 | /// |
| 130 | /// Unlike [`Array::to_data`] this consumes self, allowing it avoid unnecessary clones |
| 131 | fn into_data(self) -> ArrayData; |
| 132 | |
| 133 | /// Returns a reference to the [`DataType`] of this array. |
| 134 | /// |
| 135 | /// # Example: |
| 136 | /// |
| 137 | /// ``` |
| 138 | /// use arrow_schema::DataType; |
| 139 | /// use arrow_array::{Array, Int32Array}; |
| 140 | /// |
| 141 | /// let array = Int32Array::from(vec![1, 2, 3, 4, 5]); |
| 142 | /// |
| 143 | /// assert_eq!(*array.data_type(), DataType::Int32); |
| 144 | /// ``` |
| 145 | fn data_type(&self) -> &DataType; |
| 146 | |
| 147 | /// Returns a zero-copy slice of this array with the indicated offset and length. |
| 148 | /// |
| 149 | /// # Example: |
| 150 | /// |
| 151 | /// ``` |
| 152 | /// use arrow_array::{Array, Int32Array}; |
| 153 | /// |
| 154 | /// let array = Int32Array::from(vec![1, 2, 3, 4, 5]); |
| 155 | /// // Make slice over the values [2, 3, 4] |
| 156 | /// let array_slice = array.slice(1, 3); |
| 157 | /// |
| 158 | /// assert_eq!(&array_slice, &Int32Array::from(vec![2, 3, 4])); |
| 159 | /// ``` |
| 160 | fn slice(&self, offset: usize, length: usize) -> ArrayRef; |
| 161 | |
| 162 | /// Returns the length (i.e., number of elements) of this array. |
| 163 | /// |
| 164 | /// # Example: |
| 165 | /// |
| 166 | /// ``` |
| 167 | /// use arrow_array::{Array, Int32Array}; |
| 168 | /// |
| 169 | /// let array = Int32Array::from(vec![1, 2, 3, 4, 5]); |
| 170 | /// |
| 171 | /// assert_eq!(array.len(), 5); |
| 172 | /// ``` |
| 173 | fn len(&self) -> usize; |
| 174 | |
| 175 | /// Returns whether this array is empty. |
| 176 | /// |
| 177 | /// # Example: |
| 178 | /// |
| 179 | /// ``` |
| 180 | /// use arrow_array::{Array, Int32Array}; |
| 181 | /// |
| 182 | /// let array = Int32Array::from(vec![1, 2, 3, 4, 5]); |
| 183 | /// |
| 184 | /// assert_eq!(array.is_empty(), false); |
| 185 | /// ``` |
| 186 | fn is_empty(&self) -> bool; |
| 187 | |
| 188 | /// Shrinks the capacity of any exclusively owned buffer as much as possible |
| 189 | /// |
| 190 | /// Shared or externally allocated buffers will be ignored, and |
| 191 | /// any buffer offsets will be preserved. |
| 192 | fn shrink_to_fit(&mut self) {} |
| 193 | |
| 194 | /// Returns the offset into the underlying data used by this array(-slice). |
| 195 | /// Note that the underlying data can be shared by many arrays. |
| 196 | /// This defaults to `0`. |
| 197 | /// |
| 198 | /// # Example: |
| 199 | /// |
| 200 | /// ``` |
| 201 | /// use arrow_array::{Array, BooleanArray}; |
| 202 | /// |
| 203 | /// let array = BooleanArray::from(vec![false, false, true, true]); |
| 204 | /// let array_slice = array.slice(1, 3); |
| 205 | /// |
| 206 | /// assert_eq!(array.offset(), 0); |
| 207 | /// assert_eq!(array_slice.offset(), 1); |
| 208 | /// ``` |
| 209 | fn offset(&self) -> usize; |
| 210 | |
| 211 | /// Returns the null buffer of this array if any. |
| 212 | /// |
| 213 | /// The null buffer contains the "physical" nulls of an array, that is how |
| 214 | /// the nulls are represented in the underlying arrow format. |
| 215 | /// |
| 216 | /// The physical representation is efficient, but is sometimes non intuitive |
| 217 | /// for certain array types such as those with nullable child arrays like |
| 218 | /// [`DictionaryArray::values`], [`RunArray::values`] or [`UnionArray`], or without a |
| 219 | /// null buffer, such as [`NullArray`]. |
| 220 | /// |
| 221 | /// To determine if each element of such an array is "logically" null, |
| 222 | /// use the slower [`Array::logical_nulls`] to obtain a computed mask. |
| 223 | fn nulls(&self) -> Option<&NullBuffer>; |
| 224 | |
| 225 | /// Returns a potentially computed [`NullBuffer`] that represents the logical |
| 226 | /// null values of this array, if any. |
| 227 | /// |
| 228 | /// Logical nulls represent the values that are null in the array, |
| 229 | /// regardless of the underlying physical arrow representation. |
| 230 | /// |
| 231 | /// For most array types, this is equivalent to the "physical" nulls |
| 232 | /// returned by [`Array::nulls`]. It is different for the following cases, because which |
| 233 | /// elements are null is not encoded in a single null buffer: |
| 234 | /// |
| 235 | /// * [`DictionaryArray`] where [`DictionaryArray::values`] contains nulls |
| 236 | /// * [`RunArray`] where [`RunArray::values`] contains nulls |
| 237 | /// * [`NullArray`] where all indices are nulls |
| 238 | /// * [`UnionArray`] where the selected values contains nulls |
| 239 | /// |
| 240 | /// In these cases a logical [`NullBuffer`] will be computed, encoding the |
| 241 | /// logical nullability of these arrays, beyond what is encoded in |
| 242 | /// [`Array::nulls`] |
| 243 | fn logical_nulls(&self) -> Option<NullBuffer> { |
| 244 | self.nulls().cloned() |
| 245 | } |
| 246 | |
| 247 | /// Returns whether the element at `index` is null according to [`Array::nulls`] |
| 248 | /// |
| 249 | /// Note: For performance reasons, this method returns nullability solely as determined by the |
| 250 | /// null buffer. This difference can lead to surprising results, for example, [`NullArray::is_null`] always |
| 251 | /// returns `false` as the array lacks a null buffer. Similarly [`DictionaryArray`], [`RunArray`] and [`UnionArray`] may |
| 252 | /// encode nullability in their children. See [`Self::logical_nulls`] for more information. |
| 253 | /// |
| 254 | /// # Example: |
| 255 | /// |
| 256 | /// ``` |
| 257 | /// use arrow_array::{Array, Int32Array, NullArray}; |
| 258 | /// |
| 259 | /// let array = Int32Array::from(vec![Some(1), None]); |
| 260 | /// assert_eq!(array.is_null(0), false); |
| 261 | /// assert_eq!(array.is_null(1), true); |
| 262 | /// |
| 263 | /// // NullArrays do not have a null buffer, and therefore always |
| 264 | /// // return false for is_null. |
| 265 | /// let array = NullArray::new(1); |
| 266 | /// assert_eq!(array.is_null(0), false); |
| 267 | /// ``` |
| 268 | fn is_null(&self, index: usize) -> bool { |
| 269 | self.nulls().map(|n| n.is_null(index)).unwrap_or_default() |
| 270 | } |
| 271 | |
| 272 | /// Returns whether the element at `index` is *not* null, the |
| 273 | /// opposite of [`Self::is_null`]. |
| 274 | /// |
| 275 | /// # Example: |
| 276 | /// |
| 277 | /// ``` |
| 278 | /// use arrow_array::{Array, Int32Array}; |
| 279 | /// |
| 280 | /// let array = Int32Array::from(vec![Some(1), None]); |
| 281 | /// |
| 282 | /// assert_eq!(array.is_valid(0), true); |
| 283 | /// assert_eq!(array.is_valid(1), false); |
| 284 | /// ``` |
| 285 | fn is_valid(&self, index: usize) -> bool { |
| 286 | !self.is_null(index) |
| 287 | } |
| 288 | |
| 289 | /// Returns the total number of physical null values in this array. |
| 290 | /// |
| 291 | /// Note: this method returns the physical null count, i.e. that encoded in [`Array::nulls`], |
| 292 | /// see [`Array::logical_nulls`] for logical nullability |
| 293 | /// |
| 294 | /// # Example: |
| 295 | /// |
| 296 | /// ``` |
| 297 | /// use arrow_array::{Array, Int32Array}; |
| 298 | /// |
| 299 | /// // Construct an array with values [1, NULL, NULL] |
| 300 | /// let array = Int32Array::from(vec![Some(1), None, None]); |
| 301 | /// |
| 302 | /// assert_eq!(array.null_count(), 2); |
| 303 | /// ``` |
| 304 | fn null_count(&self) -> usize { |
| 305 | self.nulls().map(|n| n.null_count()).unwrap_or_default() |
| 306 | } |
| 307 | |
| 308 | /// Returns the total number of logical null values in this array. |
| 309 | /// |
| 310 | /// Note: this method returns the logical null count, i.e. that encoded in |
| 311 | /// [`Array::logical_nulls`]. In general this is equivalent to [`Array::null_count`] but may differ in the |
| 312 | /// presence of logical nullability, see [`Array::nulls`] and [`Array::logical_nulls`]. |
| 313 | /// |
| 314 | /// # Example: |
| 315 | /// |
| 316 | /// ``` |
| 317 | /// use arrow_array::{Array, Int32Array}; |
| 318 | /// |
| 319 | /// // Construct an array with values [1, NULL, NULL] |
| 320 | /// let array = Int32Array::from(vec![Some(1), None, None]); |
| 321 | /// |
| 322 | /// assert_eq!(array.logical_null_count(), 2); |
| 323 | /// ``` |
| 324 | fn logical_null_count(&self) -> usize { |
| 325 | self.logical_nulls() |
| 326 | .map(|n| n.null_count()) |
| 327 | .unwrap_or_default() |
| 328 | } |
| 329 | |
| 330 | /// Returns `false` if the array is guaranteed to not contain any logical nulls |
| 331 | /// |
| 332 | /// This is generally equivalent to `Array::logical_null_count() != 0` unless determining |
| 333 | /// the logical nulls is expensive, in which case this method can return true even for an |
| 334 | /// array without nulls. |
| 335 | /// |
| 336 | /// This is also generally equivalent to `Array::null_count() != 0` but may differ in the |
| 337 | /// presence of logical nullability, see [`Array::logical_null_count`] and [`Array::null_count`]. |
| 338 | /// |
| 339 | /// Implementations will return `true` unless they can cheaply prove no logical nulls |
| 340 | /// are present. For example a [`DictionaryArray`] with nullable values will still return true, |
| 341 | /// even if the nulls present in [`DictionaryArray::values`] are not referenced by any key, |
| 342 | /// and therefore would not appear in [`Array::logical_nulls`]. |
| 343 | fn is_nullable(&self) -> bool { |
| 344 | self.logical_null_count() != 0 |
| 345 | } |
| 346 | |
| 347 | /// Returns the total number of bytes of memory pointed to by this array. |
| 348 | /// The buffers store bytes in the Arrow memory format, and include the data as well as the validity map. |
| 349 | /// Note that this does not always correspond to the exact memory usage of an array, |
| 350 | /// since multiple arrays can share the same buffers or slices thereof. |
| 351 | fn get_buffer_memory_size(&self) -> usize; |
| 352 | |
| 353 | /// Returns the total number of bytes of memory occupied physically by this array. |
| 354 | /// This value will always be greater than returned by `get_buffer_memory_size()` and |
| 355 | /// includes the overhead of the data structures that contain the pointers to the various buffers. |
| 356 | fn get_array_memory_size(&self) -> usize; |
| 357 | |
| 358 | /// Claim memory used by this array in the provided memory pool. |
| 359 | /// |
| 360 | /// This recursively claims memory for: |
| 361 | /// - All data buffers in this array |
| 362 | /// - All child arrays (for nested types like List, Struct, etc.) |
| 363 | /// - The null bitmap buffer if present |
| 364 | /// |
| 365 | /// This method guarantees that the memory pool will only compute occupied memory |
| 366 | /// exactly once. For example, if this array is derived from operations like `slice`, |
| 367 | /// calling `claim` on it would not change the memory pool's usage if the underlying buffers |
| 368 | /// are already counted before. |
| 369 | /// |
| 370 | /// # Example |
| 371 | /// ``` |
| 372 | /// # use arrow_array::{Int32Array, Array}; |
| 373 | /// # use arrow_buffer::TrackingMemoryPool; |
| 374 | /// # use arrow_buffer::MemoryPool; |
| 375 | /// |
| 376 | /// let pool = TrackingMemoryPool::default(); |
| 377 | /// |
| 378 | /// let small_array = Int32Array::from(vec![1, 2, 3, 4, 5]); |
| 379 | /// let small_array_size = small_array.get_buffer_memory_size(); |
| 380 | /// |
| 381 | /// // Claim the array's memory in the pool |
| 382 | /// small_array.claim(&pool); |
| 383 | /// |
| 384 | /// // Create and claim slices of `small_array`; should not increase memory usage |
| 385 | /// let slice1 = small_array.slice(0, 2); |
| 386 | /// let slice2 = small_array.slice(2, 2); |
| 387 | /// slice1.claim(&pool); |
| 388 | /// slice2.claim(&pool); |
| 389 | /// |
| 390 | /// assert_eq!(pool.used(), small_array_size); |
| 391 | /// |
| 392 | /// // Create a `large_array` which does not derive from the original `small_array` |
| 393 | /// |
| 394 | /// let large_array = Int32Array::from((0..1000).collect::<Vec<i32>>()); |
| 395 | /// let large_array_size = large_array.get_buffer_memory_size(); |
| 396 | /// |
| 397 | /// large_array.claim(&pool); |
| 398 | /// |
| 399 | /// // Trying to claim more than once is a no-op |
| 400 | /// large_array.claim(&pool); |
| 401 | /// large_array.claim(&pool); |
| 402 | /// |
| 403 | /// assert_eq!(pool.used(), small_array_size + large_array_size); |
| 404 | /// |
| 405 | /// let sum_of_all_sizes = small_array_size + large_array_size + slice1.get_buffer_memory_size() + slice2.get_buffer_memory_size(); |
| 406 | /// |
| 407 | /// // `get_buffer_memory_size` works independently of the memory pool, so a sum of all the |
| 408 | /// // arrays in scope will always be >= the memory used reported by the memory pool. |
| 409 | /// assert_ne!(pool.used(), sum_of_all_sizes); |
| 410 | /// |
| 411 | /// // Until the final claim is dropped the buffer size remains accounted for |
| 412 | /// drop(small_array); |
| 413 | /// drop(slice1); |
| 414 | /// |
| 415 | /// assert_eq!(pool.used(), small_array_size + large_array_size); |
| 416 | /// |
| 417 | /// // Dropping this finally releases the buffer that was backing `small_array` |
| 418 | /// drop(slice2); |
| 419 | /// |
| 420 | /// assert_eq!(pool.used(), large_array_size); |
| 421 | /// ``` |
| 422 | #[cfg(feature = "pool")] |
| 423 | fn claim(&self, pool: &dyn arrow_buffer::MemoryPool) { |
| 424 | self.to_data().claim(pool) |
| 425 | } |
| 426 | } |
| 427 | |
| 428 | /// A reference-counted reference to a generic `Array` |
| 429 | pub type ArrayRef = Arc<dyn Array>; |
| 430 | |
| 431 | /// Ergonomics: Allow use of an ArrayRef as an `&dyn Array` |
| 432 | unsafe impl Array for ArrayRef { |
| 433 | fn as_any(&self) -> &dyn Any { |
| 434 | self.as_ref().as_any() |
| 435 | } |
| 436 | |
| 437 | fn to_data(&self) -> ArrayData { |
| 438 | self.as_ref().to_data() |
| 439 | } |
| 440 | |
| 441 | fn into_data(self) -> ArrayData { |
| 442 | self.to_data() |
| 443 | } |
| 444 | |
| 445 | fn data_type(&self) -> &DataType { |
| 446 | self.as_ref().data_type() |
| 447 | } |
| 448 | |
| 449 | fn slice(&self, offset: usize, length: usize) -> ArrayRef { |
| 450 | self.as_ref().slice(offset, length) |
| 451 | } |
| 452 | |
| 453 | fn len(&self) -> usize { |
| 454 | self.as_ref().len() |
| 455 | } |
| 456 | |
| 457 | fn is_empty(&self) -> bool { |
| 458 | self.as_ref().is_empty() |
| 459 | } |
| 460 | |
| 461 | /// For shared buffers, this is a no-op. |
| 462 | fn shrink_to_fit(&mut self) { |
| 463 | if let Some(slf) = Arc::get_mut(self) { |
| 464 | slf.shrink_to_fit(); |
| 465 | } else { |
| 466 | // We ignore shared buffers. |
| 467 | } |
| 468 | } |
| 469 | |
| 470 | fn offset(&self) -> usize { |
| 471 | self.as_ref().offset() |
| 472 | } |
| 473 | |
| 474 | fn nulls(&self) -> Option<&NullBuffer> { |
| 475 | self.as_ref().nulls() |
| 476 | } |
| 477 | |
| 478 | fn logical_nulls(&self) -> Option<NullBuffer> { |
| 479 | self.as_ref().logical_nulls() |
| 480 | } |
| 481 | |
| 482 | fn is_null(&self, index: usize) -> bool { |
| 483 | self.as_ref().is_null(index) |
| 484 | } |
| 485 | |
| 486 | fn is_valid(&self, index: usize) -> bool { |
| 487 | self.as_ref().is_valid(index) |
| 488 | } |
| 489 | |
| 490 | fn null_count(&self) -> usize { |
| 491 | self.as_ref().null_count() |
| 492 | } |
| 493 | |
| 494 | fn logical_null_count(&self) -> usize { |
| 495 | self.as_ref().logical_null_count() |
| 496 | } |
| 497 | |
| 498 | fn is_nullable(&self) -> bool { |
| 499 | self.as_ref().is_nullable() |
| 500 | } |
| 501 | |
| 502 | fn get_buffer_memory_size(&self) -> usize { |
| 503 | self.as_ref().get_buffer_memory_size() |
| 504 | } |
| 505 | |
| 506 | fn get_array_memory_size(&self) -> usize { |
| 507 | self.as_ref().get_array_memory_size() |
| 508 | } |
| 509 | |
| 510 | #[cfg(feature = "pool")] |
| 511 | fn claim(&self, pool: &dyn arrow_buffer::MemoryPool) { |
| 512 | self.as_ref().claim(pool) |
| 513 | } |
| 514 | } |
| 515 | |
| 516 | unsafe impl<T: Array> Array for &T { |
| 517 | fn as_any(&self) -> &dyn Any { |
| 518 | T::as_any(self) |
| 519 | } |
| 520 | |
| 521 | fn to_data(&self) -> ArrayData { |
| 522 | T::to_data(self) |
| 523 | } |
| 524 | |
| 525 | fn into_data(self) -> ArrayData { |
| 526 | self.to_data() |
| 527 | } |
| 528 | |
| 529 | fn data_type(&self) -> &DataType { |
| 530 | T::data_type(self) |
| 531 | } |
| 532 | |
| 533 | fn slice(&self, offset: usize, length: usize) -> ArrayRef { |
| 534 | T::slice(self, offset, length) |
| 535 | } |
| 536 | |
| 537 | fn len(&self) -> usize { |
| 538 | T::len(self) |
| 539 | } |
| 540 | |
| 541 | fn is_empty(&self) -> bool { |
| 542 | T::is_empty(self) |
| 543 | } |
| 544 | |
| 545 | fn offset(&self) -> usize { |
| 546 | T::offset(self) |
| 547 | } |
| 548 | |
| 549 | fn nulls(&self) -> Option<&NullBuffer> { |
| 550 | T::nulls(self) |
| 551 | } |
| 552 | |
| 553 | fn logical_nulls(&self) -> Option<NullBuffer> { |
| 554 | T::logical_nulls(self) |
| 555 | } |
| 556 | |
| 557 | fn is_null(&self, index: usize) -> bool { |
| 558 | T::is_null(self, index) |
| 559 | } |
| 560 | |
| 561 | fn is_valid(&self, index: usize) -> bool { |
| 562 | T::is_valid(self, index) |
| 563 | } |
| 564 | |
| 565 | fn null_count(&self) -> usize { |
| 566 | T::null_count(self) |
| 567 | } |
| 568 | |
| 569 | fn logical_null_count(&self) -> usize { |
| 570 | T::logical_null_count(self) |
| 571 | } |
| 572 | |
| 573 | fn is_nullable(&self) -> bool { |
| 574 | T::is_nullable(self) |
| 575 | } |
| 576 | |
| 577 | fn get_buffer_memory_size(&self) -> usize { |
| 578 | T::get_buffer_memory_size(self) |
| 579 | } |
| 580 | |
| 581 | fn get_array_memory_size(&self) -> usize { |
| 582 | T::get_array_memory_size(self) |
| 583 | } |
| 584 | |
| 585 | #[cfg(feature = "pool")] |
| 586 | fn claim(&self, pool: &dyn arrow_buffer::MemoryPool) { |
| 587 | T::claim(self, pool) |
| 588 | } |
| 589 | } |
| 590 | |
| 591 | /// A generic trait for accessing the values of an [`Array`] |
| 592 | /// |
| 593 | /// This trait helps write specialized implementations of algorithms for |
| 594 | /// different array types. Specialized implementations allow the compiler |
| 595 | /// to optimize the code for the specific array type, which can lead to |
| 596 | /// significant performance improvements. |
| 597 | /// |
| 598 | /// # Example |
| 599 | /// For example, to write three different implementations of a string length function |
| 600 | /// for [`StringArray`], [`LargeStringArray`], and [`StringViewArray`], you can write |
| 601 | /// |
| 602 | /// ``` |
| 603 | /// # use std::sync::Arc; |
| 604 | /// # use arrow_array::{ArrayAccessor, ArrayRef, ArrowPrimitiveType, OffsetSizeTrait, PrimitiveArray}; |
| 605 | /// # use arrow_buffer::ArrowNativeType; |
| 606 | /// # use arrow_array::cast::AsArray; |
| 607 | /// # use arrow_array::iterator::ArrayIter; |
| 608 | /// # use arrow_array::types::{Int32Type, Int64Type}; |
| 609 | /// # use arrow_schema::{ArrowError, DataType}; |
| 610 | /// /// This function takes a dynamically typed `ArrayRef` and calls |
| 611 | /// /// calls one of three specialized implementations |
| 612 | /// fn character_length(arg: ArrayRef) -> Result<ArrayRef, ArrowError> { |
| 613 | /// match arg.data_type() { |
| 614 | /// DataType::Utf8 => { |
| 615 | /// // downcast the ArrayRef to a StringArray and call the specialized implementation |
| 616 | /// let string_array = arg.as_string::<i32>(); |
| 617 | /// character_length_general::<Int32Type, _>(string_array) |
| 618 | /// } |
| 619 | /// DataType::LargeUtf8 => { |
| 620 | /// character_length_general::<Int64Type, _>(arg.as_string::<i64>()) |
| 621 | /// } |
| 622 | /// DataType::Utf8View => { |
| 623 | /// character_length_general::<Int32Type, _>(arg.as_string_view()) |
| 624 | /// } |
| 625 | /// _ => Err(ArrowError::InvalidArgumentError("Unsupported data type".to_string())), |
| 626 | /// } |
| 627 | /// } |
| 628 | /// |
| 629 | /// /// A generic implementation of the character_length function |
| 630 | /// /// This function uses the `ArrayAccessor` trait to access the values of the array |
| 631 | /// /// so the compiler can generated specialized implementations for different array types |
| 632 | /// /// |
| 633 | /// /// Returns a new array with the length of each string in the input array |
| 634 | /// /// * Int32Array for Utf8 and Utf8View arrays (lengths are 32-bit integers) |
| 635 | /// /// * Int64Array for LargeUtf8 arrays (lengths are 64-bit integers) |
| 636 | /// /// |
| 637 | /// /// This is generic on the type of the primitive array (different string arrays have |
| 638 | /// /// different lengths) and the type of the array accessor (different string arrays |
| 639 | /// /// have different ways to access the values) |
| 640 | /// fn character_length_general<'a, T: ArrowPrimitiveType, V: ArrayAccessor<Item = &'a str>>( |
| 641 | /// array: V, |
| 642 | /// ) -> Result<ArrayRef, ArrowError> |
| 643 | /// where |
| 644 | /// T::Native: OffsetSizeTrait, |
| 645 | /// { |
| 646 | /// let iter = ArrayIter::new(array); |
| 647 | /// // Create a Int32Array / Int64Array with the length of each string |
| 648 | /// let result = iter |
| 649 | /// .map(|string| { |
| 650 | /// string.map(|string: &str| { |
| 651 | /// T::Native::from_usize(string.chars().count()) |
| 652 | /// .expect("should not fail as string.chars will always return integer") |
| 653 | /// }) |
| 654 | /// }) |
| 655 | /// .collect::<PrimitiveArray<T>>(); |
| 656 | /// |
| 657 | /// /// Return the result as a new ArrayRef (dynamically typed) |
| 658 | /// Ok(Arc::new(result) as ArrayRef) |
| 659 | /// } |
| 660 | /// ``` |
| 661 | /// |
| 662 | /// # Validity |
| 663 | /// |
| 664 | /// An [`ArrayAccessor`] must always return a well-defined value for an index |
| 665 | /// that is within the bounds `0..Array::len`, including for null indexes where |
| 666 | /// [`Array::is_null`] is true. |
| 667 | /// |
| 668 | /// The value at null indexes is unspecified, and implementations must not rely |
| 669 | /// on a specific value such as [`Default::default`] being returned, however, it |
| 670 | /// must not be undefined |
| 671 | pub trait ArrayAccessor: Array { |
| 672 | /// The Arrow type of the element being accessed. |
| 673 | type Item: Send + Sync; |
| 674 | |
| 675 | /// Returns the element at index `i` |
| 676 | /// # Panics |
| 677 | /// Panics if the value is outside the bounds of the array |
| 678 | fn value(&self, index: usize) -> Self::Item; |
| 679 | |
| 680 | /// Returns the element at index `i` |
| 681 | /// # Safety |
| 682 | /// Caller is responsible for ensuring that the index is within the bounds of the array |
| 683 | unsafe fn value_unchecked(&self, index: usize) -> Self::Item; |
| 684 | } |
| 685 | |
| 686 | /// A trait for Arrow String Arrays, currently three types are supported: |
| 687 | /// - `StringArray` |
| 688 | /// - `LargeStringArray` |
| 689 | /// - `StringViewArray` |
| 690 | /// |
| 691 | /// This trait helps to abstract over the different types of string arrays |
| 692 | /// so that we don't need to duplicate the implementation for each type. |
| 693 | pub trait StringArrayType<'a>: ArrayAccessor<Item = &'a str> + Sized { |
| 694 | /// Returns true if all data within this string array is ASCII |
| 695 | fn is_ascii(&self) -> bool; |
| 696 | |
| 697 | /// Constructs a new iterator |
| 698 | fn iter(&self) -> ArrayIter<Self>; |
| 699 | } |
| 700 | |
| 701 | impl<'a, O: OffsetSizeTrait> StringArrayType<'a> for &'a GenericStringArray<O> { |
| 702 | fn is_ascii(&self) -> bool { |
| 703 | GenericStringArray::<O>::is_ascii(self) |
| 704 | } |
| 705 | |
| 706 | fn iter(&self) -> ArrayIter<Self> { |
| 707 | GenericStringArray::<O>::iter(self) |
| 708 | } |
| 709 | } |
| 710 | impl<'a> StringArrayType<'a> for &'a StringViewArray { |
| 711 | fn is_ascii(&self) -> bool { |
| 712 | StringViewArray::is_ascii(self) |
| 713 | } |
| 714 | |
| 715 | fn iter(&self) -> ArrayIter<Self> { |
| 716 | StringViewArray::iter(self) |
| 717 | } |
| 718 | } |
| 719 | |
| 720 | /// A trait for Arrow Binary Arrays, currently four types are supported: |
| 721 | /// - `BinaryArray` |
| 722 | /// - `LargeBinaryArray` |
| 723 | /// - `BinaryViewArray` |
| 724 | /// - `FixedSizeBinaryArray` |
| 725 | /// |
| 726 | /// This trait helps to abstract over the different types of binary arrays |
| 727 | /// so that we don't need to duplicate the implementation for each type. |
| 728 | pub trait BinaryArrayType<'a>: ArrayAccessor<Item = &'a [u8]> + Sized { |
| 729 | /// Constructs a new iterator |
| 730 | fn iter(&self) -> ArrayIter<Self>; |
| 731 | } |
| 732 | |
| 733 | impl<'a, O: OffsetSizeTrait> BinaryArrayType<'a> for &'a GenericBinaryArray<O> { |
| 734 | fn iter(&self) -> ArrayIter<Self> { |
| 735 | GenericBinaryArray::<O>::iter(self) |
| 736 | } |
| 737 | } |
| 738 | impl<'a> BinaryArrayType<'a> for &'a BinaryViewArray { |
| 739 | fn iter(&self) -> ArrayIter<Self> { |
| 740 | BinaryViewArray::iter(self) |
| 741 | } |
| 742 | } |
| 743 | impl<'a> BinaryArrayType<'a> for &'a FixedSizeBinaryArray { |
| 744 | fn iter(&self) -> ArrayIter<Self> { |
| 745 | FixedSizeBinaryArray::iter(self) |
| 746 | } |
| 747 | } |
| 748 | |
| 749 | /// A trait for Arrow list-like arrays, abstracting over |
| 750 | /// [`GenericListArray`], [`GenericListViewArray`], and [`FixedSizeListArray`]. |
| 751 | /// |
| 752 | /// This trait provides a uniform interface for accessing the child values and |
| 753 | /// computing the element range for a given index, regardless of the underlying |
| 754 | /// list layout (offsets, offsets+sizes, or fixed-size). |
| 755 | pub trait ListLikeArray: Array { |
| 756 | /// Returns the child values array. |
| 757 | fn values(&self) -> &ArrayRef; |
| 758 | |
| 759 | /// Returns the start and end indices into the values array for the list |
| 760 | /// element at `index`. |
| 761 | fn element_range(&self, index: usize) -> std::ops::Range<usize>; |
| 762 | } |
| 763 | |
| 764 | impl PartialEq for dyn Array + '_ { |
| 765 | fn eq(&self, other: &Self) -> bool { |
| 766 | self.to_data().eq(&other.to_data()) |
| 767 | } |
| 768 | } |
| 769 | |
| 770 | impl<T: Array> PartialEq<T> for dyn Array + '_ { |
| 771 | fn eq(&self, other: &T) -> bool { |
| 772 | self.to_data().eq(&other.to_data()) |
| 773 | } |
| 774 | } |
| 775 | |
| 776 | impl PartialEq for NullArray { |
| 777 | fn eq(&self, other: &NullArray) -> bool { |
| 778 | self.to_data().eq(&other.to_data()) |
| 779 | } |
| 780 | } |
| 781 | |
| 782 | impl<T: ArrowPrimitiveType> PartialEq for PrimitiveArray<T> { |
| 783 | fn eq(&self, other: &PrimitiveArray<T>) -> bool { |
| 784 | self.to_data().eq(&other.to_data()) |
| 785 | } |
| 786 | } |
| 787 | |
| 788 | impl<K: ArrowDictionaryKeyType> PartialEq for DictionaryArray<K> { |
| 789 | fn eq(&self, other: &Self) -> bool { |
| 790 | self.to_data().eq(&other.to_data()) |
| 791 | } |
| 792 | } |
| 793 | |
| 794 | impl PartialEq for BooleanArray { |
| 795 | fn eq(&self, other: &BooleanArray) -> bool { |
| 796 | self.to_data().eq(&other.to_data()) |
| 797 | } |
| 798 | } |
| 799 | |
| 800 | impl<OffsetSize: OffsetSizeTrait> PartialEq for GenericStringArray<OffsetSize> { |
| 801 | fn eq(&self, other: &Self) -> bool { |
| 802 | self.to_data().eq(&other.to_data()) |
| 803 | } |
| 804 | } |
| 805 | |
| 806 | impl<OffsetSize: OffsetSizeTrait> PartialEq for GenericBinaryArray<OffsetSize> { |
| 807 | fn eq(&self, other: &Self) -> bool { |
| 808 | self.to_data().eq(&other.to_data()) |
| 809 | } |
| 810 | } |
| 811 | |
| 812 | impl PartialEq for FixedSizeBinaryArray { |
| 813 | fn eq(&self, other: &Self) -> bool { |
| 814 | self.to_data().eq(&other.to_data()) |
| 815 | } |
| 816 | } |
| 817 | |
| 818 | impl<OffsetSize: OffsetSizeTrait> PartialEq for GenericListArray<OffsetSize> { |
| 819 | fn eq(&self, other: &Self) -> bool { |
| 820 | self.to_data().eq(&other.to_data()) |
| 821 | } |
| 822 | } |
| 823 | |
| 824 | impl<OffsetSize: OffsetSizeTrait> PartialEq for GenericListViewArray<OffsetSize> { |
| 825 | fn eq(&self, other: &Self) -> bool { |
| 826 | self.to_data().eq(&other.to_data()) |
| 827 | } |
| 828 | } |
| 829 | |
| 830 | impl PartialEq for MapArray { |
| 831 | fn eq(&self, other: &Self) -> bool { |
| 832 | self.to_data().eq(&other.to_data()) |
| 833 | } |
| 834 | } |
| 835 | |
| 836 | impl PartialEq for FixedSizeListArray { |
| 837 | fn eq(&self, other: &Self) -> bool { |
| 838 | self.to_data().eq(&other.to_data()) |
| 839 | } |
| 840 | } |
| 841 | |
| 842 | impl PartialEq for StructArray { |
| 843 | fn eq(&self, other: &Self) -> bool { |
| 844 | self.to_data().eq(&other.to_data()) |
| 845 | } |
| 846 | } |
| 847 | |
| 848 | impl<T: ByteViewType + ?Sized> PartialEq for GenericByteViewArray<T> { |
| 849 | fn eq(&self, other: &Self) -> bool { |
| 850 | self.to_data().eq(&other.to_data()) |
| 851 | } |
| 852 | } |
| 853 | |
| 854 | impl<R: RunEndIndexType> PartialEq for RunArray<R> { |
| 855 | fn eq(&self, other: &Self) -> bool { |
| 856 | self.to_data().eq(&other.to_data()) |
| 857 | } |
| 858 | } |
| 859 | |
| 860 | /// Constructs an [`ArrayRef`] from an [`ArrayData`]. |
| 861 | /// |
| 862 | /// # Notes: |
| 863 | /// |
| 864 | /// It is more efficient to directly construct the concrete array type rather |
| 865 | /// than using this function as creating an `ArrayData` requires at least one |
| 866 | /// additional allocation (the Vec of buffers). |
| 867 | /// |
| 868 | /// # Example: |
| 869 | /// ``` |
| 870 | /// # use std::sync::Arc; |
| 871 | /// # use arrow_data::ArrayData; |
| 872 | /// # use arrow_array::{make_array, ArrayRef, Int32Array}; |
| 873 | /// # use arrow_buffer::{Buffer, ScalarBuffer}; |
| 874 | /// # use arrow_schema::DataType; |
| 875 | /// // Create an Int32Array with values [1, 2, 3] |
| 876 | /// let values_buffer = Buffer::from_slice_ref(&[1, 2, 3]); |
| 877 | /// // ArrayData can be constructed using ArrayDataBuilder |
| 878 | /// let builder = ArrayData::builder(DataType::Int32) |
| 879 | /// .len(3) |
| 880 | /// .add_buffer(values_buffer.clone()); |
| 881 | /// let array_data = builder.build().unwrap(); |
| 882 | /// // Create the ArrayRef from the ArrayData |
| 883 | /// let array = make_array(array_data); |
| 884 | /// |
| 885 | /// // It is equivalent to directly constructing the Int32Array |
| 886 | /// let scalar_buffer = ScalarBuffer::from(values_buffer); |
| 887 | /// let int32_array: ArrayRef = Arc::new(Int32Array::new(scalar_buffer, None)); |
| 888 | /// assert_eq!(&array, &int32_array); |
| 889 | /// ``` |
| 890 | pub fn make_array(data: ArrayData) -> ArrayRef { |
| 891 | match data.data_type() { |
| 892 | DataType::Boolean => Arc::new(BooleanArray::from(data)) as ArrayRef, |
| 893 | DataType::Int8 => Arc::new(Int8Array::from(data)) as ArrayRef, |
| 894 | DataType::Int16 => Arc::new(Int16Array::from(data)) as ArrayRef, |
| 895 | DataType::Int32 => Arc::new(Int32Array::from(data)) as ArrayRef, |
| 896 | DataType::Int64 => Arc::new(Int64Array::from(data)) as ArrayRef, |
| 897 | DataType::UInt8 => Arc::new(UInt8Array::from(data)) as ArrayRef, |
| 898 | DataType::UInt16 => Arc::new(UInt16Array::from(data)) as ArrayRef, |
| 899 | DataType::UInt32 => Arc::new(UInt32Array::from(data)) as ArrayRef, |
| 900 | DataType::UInt64 => Arc::new(UInt64Array::from(data)) as ArrayRef, |
| 901 | DataType::Float16 => Arc::new(Float16Array::from(data)) as ArrayRef, |
| 902 | DataType::Float32 => Arc::new(Float32Array::from(data)) as ArrayRef, |
| 903 | DataType::Float64 => Arc::new(Float64Array::from(data)) as ArrayRef, |
| 904 | DataType::Date32 => Arc::new(Date32Array::from(data)) as ArrayRef, |
| 905 | DataType::Date64 => Arc::new(Date64Array::from(data)) as ArrayRef, |
| 906 | DataType::Time32(TimeUnit::Second) => Arc::new(Time32SecondArray::from(data)) as ArrayRef, |
| 907 | DataType::Time32(TimeUnit::Millisecond) => { |
| 908 | Arc::new(Time32MillisecondArray::from(data)) as ArrayRef |
| 909 | } |
| 910 | DataType::Time64(TimeUnit::Microsecond) => { |
| 911 | Arc::new(Time64MicrosecondArray::from(data)) as ArrayRef |
| 912 | } |
| 913 | DataType::Time64(TimeUnit::Nanosecond) => { |
| 914 | Arc::new(Time64NanosecondArray::from(data)) as ArrayRef |
| 915 | } |
| 916 | DataType::Timestamp(TimeUnit::Second, _) => { |
| 917 | Arc::new(TimestampSecondArray::from(data)) as ArrayRef |
| 918 | } |
| 919 | DataType::Timestamp(TimeUnit::Millisecond, _) => { |
| 920 | Arc::new(TimestampMillisecondArray::from(data)) as ArrayRef |
| 921 | } |
| 922 | DataType::Timestamp(TimeUnit::Microsecond, _) => { |
| 923 | Arc::new(TimestampMicrosecondArray::from(data)) as ArrayRef |
| 924 | } |
| 925 | DataType::Timestamp(TimeUnit::Nanosecond, _) => { |
| 926 | Arc::new(TimestampNanosecondArray::from(data)) as ArrayRef |
| 927 | } |
| 928 | DataType::Interval(IntervalUnit::YearMonth) => { |
| 929 | Arc::new(IntervalYearMonthArray::from(data)) as ArrayRef |
| 930 | } |
| 931 | DataType::Interval(IntervalUnit::DayTime) => { |
| 932 | Arc::new(IntervalDayTimeArray::from(data)) as ArrayRef |
| 933 | } |
| 934 | DataType::Interval(IntervalUnit::MonthDayNano) => { |
| 935 | Arc::new(IntervalMonthDayNanoArray::from(data)) as ArrayRef |
| 936 | } |
| 937 | DataType::Duration(TimeUnit::Second) => { |
| 938 | Arc::new(DurationSecondArray::from(data)) as ArrayRef |
| 939 | } |
| 940 | DataType::Duration(TimeUnit::Millisecond) => { |
| 941 | Arc::new(DurationMillisecondArray::from(data)) as ArrayRef |
| 942 | } |
| 943 | DataType::Duration(TimeUnit::Microsecond) => { |
| 944 | Arc::new(DurationMicrosecondArray::from(data)) as ArrayRef |
| 945 | } |
| 946 | DataType::Duration(TimeUnit::Nanosecond) => { |
| 947 | Arc::new(DurationNanosecondArray::from(data)) as ArrayRef |
| 948 | } |
| 949 | DataType::Binary => Arc::new(BinaryArray::from(data)) as ArrayRef, |
| 950 | DataType::LargeBinary => Arc::new(LargeBinaryArray::from(data)) as ArrayRef, |
| 951 | DataType::FixedSizeBinary(_) => Arc::new(FixedSizeBinaryArray::from(data)) as ArrayRef, |
| 952 | DataType::BinaryView => Arc::new(BinaryViewArray::from(data)) as ArrayRef, |
| 953 | DataType::Utf8 => Arc::new(StringArray::from(data)) as ArrayRef, |
| 954 | DataType::LargeUtf8 => Arc::new(LargeStringArray::from(data)) as ArrayRef, |
| 955 | DataType::Utf8View => Arc::new(StringViewArray::from(data)) as ArrayRef, |
| 956 | DataType::List(_) => Arc::new(ListArray::from(data)) as ArrayRef, |
| 957 | DataType::LargeList(_) => Arc::new(LargeListArray::from(data)) as ArrayRef, |
| 958 | DataType::ListView(_) => Arc::new(ListViewArray::from(data)) as ArrayRef, |
| 959 | DataType::LargeListView(_) => Arc::new(LargeListViewArray::from(data)) as ArrayRef, |
| 960 | DataType::Struct(_) => Arc::new(StructArray::from(data)) as ArrayRef, |
| 961 | DataType::Map(_, _) => Arc::new(MapArray::from(data)) as ArrayRef, |
| 962 | DataType::Union(_, _) => Arc::new(UnionArray::from(data)) as ArrayRef, |
| 963 | DataType::FixedSizeList(_, _) => Arc::new(FixedSizeListArray::from(data)) as ArrayRef, |
| 964 | DataType::Dictionary(key_type, _) => match key_type.as_ref() { |
| 965 | DataType::Int8 => Arc::new(DictionaryArray::<Int8Type>::from(data)) as ArrayRef, |
| 966 | DataType::Int16 => Arc::new(DictionaryArray::<Int16Type>::from(data)) as ArrayRef, |
| 967 | DataType::Int32 => Arc::new(DictionaryArray::<Int32Type>::from(data)) as ArrayRef, |
| 968 | DataType::Int64 => Arc::new(DictionaryArray::<Int64Type>::from(data)) as ArrayRef, |
| 969 | DataType::UInt8 => Arc::new(DictionaryArray::<UInt8Type>::from(data)) as ArrayRef, |
| 970 | DataType::UInt16 => Arc::new(DictionaryArray::<UInt16Type>::from(data)) as ArrayRef, |
| 971 | DataType::UInt32 => Arc::new(DictionaryArray::<UInt32Type>::from(data)) as ArrayRef, |
| 972 | DataType::UInt64 => Arc::new(DictionaryArray::<UInt64Type>::from(data)) as ArrayRef, |
| 973 | dt => unimplemented!("Unexpected dictionary key type {dt}"), |
| 974 | }, |
| 975 | DataType::RunEndEncoded(run_ends_type, _) => match run_ends_type.data_type() { |
| 976 | DataType::Int16 => Arc::new(RunArray::<Int16Type>::from(data)) as ArrayRef, |
| 977 | DataType::Int32 => Arc::new(RunArray::<Int32Type>::from(data)) as ArrayRef, |
| 978 | DataType::Int64 => Arc::new(RunArray::<Int64Type>::from(data)) as ArrayRef, |
| 979 | dt => unimplemented!("Unexpected data type for run_ends array {dt}"), |
| 980 | }, |
| 981 | DataType::Null => Arc::new(NullArray::from(data)) as ArrayRef, |
| 982 | DataType::Decimal32(_, _) => Arc::new(Decimal32Array::from(data)) as ArrayRef, |
| 983 | DataType::Decimal64(_, _) => Arc::new(Decimal64Array::from(data)) as ArrayRef, |
| 984 | DataType::Decimal128(_, _) => Arc::new(Decimal128Array::from(data)) as ArrayRef, |
| 985 | DataType::Decimal256(_, _) => Arc::new(Decimal256Array::from(data)) as ArrayRef, |
| 986 | dt => unimplemented!("Unexpected data type {dt}"), |
| 987 | } |
| 988 | } |
| 989 | |
| 990 | /// Creates a new empty array |
| 991 | /// |
| 992 | /// ``` |
| 993 | /// use std::sync::Arc; |
| 994 | /// use arrow_schema::DataType; |
| 995 | /// use arrow_array::{ArrayRef, Int32Array, new_empty_array}; |
| 996 | /// |
| 997 | /// let empty_array = new_empty_array(&DataType::Int32); |
| 998 | /// let array: ArrayRef = Arc::new(Int32Array::from(vec![] as Vec<i32>)); |
| 999 | /// |
| 1000 | /// assert_eq!(&array, &empty_array); |
| 1001 | /// ``` |
| 1002 | pub fn new_empty_array(data_type: &DataType) -> ArrayRef { |
| 1003 | let data = ArrayData::new_empty(data_type); |
| 1004 | make_array(data) |
| 1005 | } |
| 1006 | |
| 1007 | /// Creates a new array of `data_type` of length `length` filled |
| 1008 | /// entirely of `NULL` values |
| 1009 | /// |
| 1010 | /// ``` |
| 1011 | /// use std::sync::Arc; |
| 1012 | /// use arrow_schema::DataType; |
| 1013 | /// use arrow_array::{ArrayRef, Int32Array, new_null_array}; |
| 1014 | /// |
| 1015 | /// let null_array = new_null_array(&DataType::Int32, 3); |
| 1016 | /// let array: ArrayRef = Arc::new(Int32Array::from(vec![None, None, None])); |
| 1017 | /// |
| 1018 | /// assert_eq!(&array, &null_array); |
| 1019 | /// ``` |
| 1020 | pub fn new_null_array(data_type: &DataType, length: usize) -> ArrayRef { |
| 1021 | make_array(ArrayData::new_null(data_type, length)) |
| 1022 | } |
| 1023 | |
| 1024 | /// Helper function that creates an [`OffsetBuffer`] from a buffer and array offset/ length |
| 1025 | /// |
| 1026 | /// # Safety |
| 1027 | /// |
| 1028 | /// - buffer must contain valid arrow offsets ( [`OffsetBuffer`] ) for the |
| 1029 | /// given length and offset. |
| 1030 | unsafe fn get_offsets_from_buffer<O: ArrowNativeType>( |
| 1031 | buffer: Buffer, |
| 1032 | offset: usize, |
| 1033 | len: usize, |
| 1034 | ) -> OffsetBuffer<O> { |
| 1035 | if len == 0 && buffer.is_empty() { |
| 1036 | return OffsetBuffer::new_empty(); |
| 1037 | } |
| 1038 | |
| 1039 | let scalar_buffer = ScalarBuffer::new(buffer, offset, len + 1); |
| 1040 | // Safety: |
| 1041 | // Arguments were valid |
| 1042 | unsafe { OffsetBuffer::new_unchecked(scalar_buffer) } |
| 1043 | } |
| 1044 | |
| 1045 | /// Helper function for printing potentially long arrays. |
| 1046 | fn print_long_array<A, F>(array: &A, f: &mut std::fmt::Formatter, print_item: F) -> std::fmt::Result |
| 1047 | where |
| 1048 | A: Array, |
| 1049 | F: Fn(&A, usize, &mut std::fmt::Formatter) -> std::fmt::Result, |
| 1050 | { |
| 1051 | let head = std::cmp::min(10, array.len()); |
| 1052 | |
| 1053 | for i in 0..head { |
| 1054 | if array.is_null(i) { |
| 1055 | writeln!(f, " null,")?; |
| 1056 | } else { |
| 1057 | write!(f, " ")?; |
| 1058 | print_item(array, i, f)?; |
| 1059 | writeln!(f, ",")?; |
| 1060 | } |
| 1061 | } |
| 1062 | if array.len() > 10 { |
| 1063 | if array.len() > 20 { |
| 1064 | writeln!(f, " ...{} elements...,", array.len() - 20)?; |
| 1065 | } |
| 1066 | |
| 1067 | let tail = std::cmp::max(head, array.len() - 10); |
| 1068 | |
| 1069 | for i in tail..array.len() { |
| 1070 | if array.is_null(i) { |
| 1071 | writeln!(f, " null,")?; |
| 1072 | } else { |
| 1073 | write!(f, " ")?; |
| 1074 | print_item(array, i, f)?; |
| 1075 | writeln!(f, ",")?; |
| 1076 | } |
| 1077 | } |
| 1078 | } |
| 1079 | Ok(()) |
| 1080 | } |
| 1081 | |
| 1082 | #[cfg(test)] |
| 1083 | mod tests { |
| 1084 | use super::*; |
| 1085 | use crate::cast::{as_union_array, downcast_array}; |
| 1086 | use crate::downcast_run_array; |
| 1087 | use arrow_buffer::MutableBuffer; |
| 1088 | use arrow_schema::{Field, Fields, UnionFields, UnionMode}; |
| 1089 | |
| 1090 | #[test] |
| 1091 | fn test_empty_primitive() { |
| 1092 | let array = new_empty_array(&DataType::Int32); |
| 1093 | let a = array.as_any().downcast_ref::<Int32Array>().unwrap(); |
| 1094 | assert_eq!(a.len(), 0); |
| 1095 | let expected: &[i32] = &[]; |
| 1096 | assert_eq!(a.values(), expected); |
| 1097 | } |
| 1098 | |
| 1099 | #[test] |
| 1100 | fn test_empty_variable_sized() { |
| 1101 | let array = new_empty_array(&DataType::Utf8); |
| 1102 | let a = array.as_any().downcast_ref::<StringArray>().unwrap(); |
| 1103 | assert_eq!(a.len(), 0); |
| 1104 | assert_eq!(a.value_offsets()[0], 0i32); |
| 1105 | } |
| 1106 | |
| 1107 | #[test] |
| 1108 | fn test_empty_list_primitive() { |
| 1109 | let data_type = DataType::List(Arc::new(Field::new_list_field(DataType::Int32, false))); |
| 1110 | let array = new_empty_array(&data_type); |
| 1111 | let a = array.as_any().downcast_ref::<ListArray>().unwrap(); |
| 1112 | assert_eq!(a.len(), 0); |
| 1113 | assert_eq!(a.value_offsets()[0], 0i32); |
| 1114 | } |
| 1115 | |
| 1116 | #[test] |
| 1117 | fn test_null_boolean() { |
| 1118 | let array = new_null_array(&DataType::Boolean, 9); |
| 1119 | let a = array.as_any().downcast_ref::<BooleanArray>().unwrap(); |
| 1120 | assert_eq!(a.len(), 9); |
| 1121 | for i in 0..9 { |
| 1122 | assert!(a.is_null(i)); |
| 1123 | } |
| 1124 | } |
| 1125 | |
| 1126 | #[test] |
| 1127 | fn test_null_primitive() { |
| 1128 | let array = new_null_array(&DataType::Int32, 9); |
| 1129 | let a = array.as_any().downcast_ref::<Int32Array>().unwrap(); |
| 1130 | assert_eq!(a.len(), 9); |
| 1131 | for i in 0..9 { |
| 1132 | assert!(a.is_null(i)); |
| 1133 | } |
| 1134 | } |
| 1135 | |
| 1136 | #[test] |
| 1137 | fn test_null_struct() { |
| 1138 | // It is possible to create a null struct containing a non-nullable child |
| 1139 | // see https://github.com/apache/arrow-rs/pull/3244 for details |
| 1140 | let struct_type = DataType::Struct(vec![Field::new("data", DataType::Int64, false)].into()); |
| 1141 | let array = new_null_array(&struct_type, 9); |
| 1142 | |
| 1143 | let a = array.as_any().downcast_ref::<StructArray>().unwrap(); |
| 1144 | assert_eq!(a.len(), 9); |
| 1145 | assert_eq!(a.column(0).len(), 9); |
| 1146 | for i in 0..9 { |
| 1147 | assert!(a.is_null(i)); |
| 1148 | } |
| 1149 | |
| 1150 | // Make sure we can slice the resulting array. |
| 1151 | a.slice(0, 5); |
| 1152 | } |
| 1153 | |
| 1154 | #[test] |
| 1155 | fn test_null_variable_sized() { |
| 1156 | let array = new_null_array(&DataType::Utf8, 9); |
| 1157 | let a = array.as_any().downcast_ref::<StringArray>().unwrap(); |
| 1158 | assert_eq!(a.len(), 9); |
| 1159 | assert_eq!(a.value_offsets()[9], 0i32); |
| 1160 | for i in 0..9 { |
| 1161 | assert!(a.is_null(i)); |
| 1162 | } |
| 1163 | } |
| 1164 | |
| 1165 | #[test] |
| 1166 | fn test_null_list_primitive() { |
| 1167 | let data_type = DataType::List(Arc::new(Field::new_list_field(DataType::Int32, true))); |
| 1168 | let array = new_null_array(&data_type, 9); |
| 1169 | let a = array.as_any().downcast_ref::<ListArray>().unwrap(); |
| 1170 | assert_eq!(a.len(), 9); |
| 1171 | assert_eq!(a.value_offsets()[9], 0i32); |
| 1172 | for i in 0..9 { |
| 1173 | assert!(a.is_null(i)); |
| 1174 | } |
| 1175 | } |
| 1176 | |
| 1177 | #[test] |
| 1178 | fn test_null_map() { |
| 1179 | let data_type = DataType::Map( |
| 1180 | Arc::new(Field::new( |
| 1181 | "entry", |
| 1182 | DataType::Struct(Fields::from(vec![ |
| 1183 | Field::new("key", DataType::Utf8, false), |
| 1184 | Field::new("value", DataType::Int32, true), |
| 1185 | ])), |
| 1186 | false, |
| 1187 | )), |
| 1188 | false, |
| 1189 | ); |
| 1190 | let array = new_null_array(&data_type, 9); |
| 1191 | let a = array.as_any().downcast_ref::<MapArray>().unwrap(); |
| 1192 | assert_eq!(a.len(), 9); |
| 1193 | assert_eq!(a.value_offsets()[9], 0i32); |
| 1194 | for i in 0..9 { |
| 1195 | assert!(a.is_null(i)); |
| 1196 | } |
| 1197 | } |
| 1198 | |
| 1199 | #[test] |
| 1200 | fn test_null_dictionary() { |
| 1201 | let values = |
| 1202 | vec![None, None, None, None, None, None, None, None, None] as Vec<Option<&str>>; |
| 1203 | |
| 1204 | let array: DictionaryArray<Int8Type> = values.into_iter().collect(); |
| 1205 | let array = Arc::new(array) as ArrayRef; |
| 1206 | |
| 1207 | let null_array = new_null_array(array.data_type(), 9); |
| 1208 | assert_eq!(&array, &null_array); |
| 1209 | assert_eq!( |
| 1210 | array.to_data().buffers()[0].len(), |
| 1211 | null_array.to_data().buffers()[0].len() |
| 1212 | ); |
| 1213 | } |
| 1214 | |
| 1215 | #[test] |
| 1216 | fn test_null_union() { |
| 1217 | for mode in [UnionMode::Sparse, UnionMode::Dense] { |
| 1218 | let data_type = DataType::Union( |
| 1219 | UnionFields::try_new( |
| 1220 | vec![2, 1], |
| 1221 | vec![ |
| 1222 | Field::new("foo", DataType::Int32, true), |
| 1223 | Field::new("bar", DataType::Int64, true), |
| 1224 | ], |
| 1225 | ) |
| 1226 | .unwrap(), |
| 1227 | mode, |
| 1228 | ); |
| 1229 | let array = new_null_array(&data_type, 4); |
| 1230 | |
| 1231 | let array = as_union_array(array.as_ref()); |
| 1232 | assert_eq!(array.len(), 4); |
| 1233 | assert_eq!(array.null_count(), 0); |
| 1234 | assert_eq!(array.logical_null_count(), 4); |
| 1235 | |
| 1236 | for i in 0..4 { |
| 1237 | let a = array.value(i); |
| 1238 | assert_eq!(a.len(), 1); |
| 1239 | assert_eq!(a.null_count(), 1); |
| 1240 | assert_eq!(a.logical_null_count(), 1); |
| 1241 | assert!(a.is_null(0)) |
| 1242 | } |
| 1243 | |
| 1244 | array.to_data().validate_full().unwrap(); |
| 1245 | } |
| 1246 | } |
| 1247 | |
| 1248 | #[test] |
| 1249 | #[allow(unused_parens)] |
| 1250 | fn test_null_runs() { |
| 1251 | for r in [DataType::Int16, DataType::Int32, DataType::Int64] { |
| 1252 | let data_type = DataType::RunEndEncoded( |
| 1253 | Arc::new(Field::new("run_ends", r, false)), |
| 1254 | Arc::new(Field::new("values", DataType::Utf8, true)), |
| 1255 | ); |
| 1256 | |
| 1257 | let array = new_null_array(&data_type, 4); |
| 1258 | let array = array.as_ref(); |
| 1259 | |
| 1260 | downcast_run_array! { |
| 1261 | array => { |
| 1262 | assert_eq!(array.len(), 4); |
| 1263 | assert_eq!(array.null_count(), 0); |
| 1264 | assert_eq!(array.logical_null_count(), 4); |
| 1265 | assert_eq!(array.values().len(), 1); |
| 1266 | assert_eq!(array.values().null_count(), 1); |
| 1267 | assert_eq!(array.run_ends().len(), 4); |
| 1268 | assert_eq!(array.run_ends().values(), &[4]); |
| 1269 | |
| 1270 | let idx = array.get_physical_indices(&[0, 1, 2, 3]).unwrap(); |
| 1271 | assert_eq!(idx, &[0,0,0,0]); |
| 1272 | } |
| 1273 | d => unreachable!("{d}") |
| 1274 | } |
| 1275 | } |
| 1276 | } |
| 1277 | |
| 1278 | #[test] |
| 1279 | fn test_null_fixed_size_binary() { |
| 1280 | for size in [1, 2, 7] { |
| 1281 | let array = new_null_array(&DataType::FixedSizeBinary(size), 6); |
| 1282 | let array = array |
| 1283 | .as_ref() |
| 1284 | .as_any() |
| 1285 | .downcast_ref::<FixedSizeBinaryArray>() |
| 1286 | .unwrap(); |
| 1287 | |
| 1288 | assert_eq!(array.len(), 6); |
| 1289 | assert_eq!(array.null_count(), 6); |
| 1290 | assert_eq!(array.logical_null_count(), 6); |
| 1291 | array.iter().for_each(|x| assert!(x.is_none())); |
| 1292 | } |
| 1293 | } |
| 1294 | |
| 1295 | #[test] |
| 1296 | fn test_memory_size_null() { |
| 1297 | let null_arr = NullArray::new(32); |
| 1298 | |
| 1299 | assert_eq!(0, null_arr.get_buffer_memory_size()); |
| 1300 | assert_eq!( |
| 1301 | std::mem::size_of::<usize>(), |
| 1302 | null_arr.get_array_memory_size() |
| 1303 | ); |
| 1304 | } |
| 1305 | |
| 1306 | #[test] |
| 1307 | fn test_memory_size_primitive() { |
| 1308 | let arr = PrimitiveArray::<Int64Type>::from_iter_values(0..128); |
| 1309 | let empty = PrimitiveArray::<Int64Type>::from(ArrayData::new_empty(arr.data_type())); |
| 1310 | |
| 1311 | // subtract empty array to avoid magic numbers for the size of additional fields |
| 1312 | assert_eq!( |
| 1313 | arr.get_array_memory_size() - empty.get_array_memory_size(), |
| 1314 | 128 * std::mem::size_of::<i64>() |
| 1315 | ); |
| 1316 | } |
| 1317 | |
| 1318 | #[test] |
| 1319 | fn test_memory_size_primitive_sliced() { |
| 1320 | let arr = PrimitiveArray::<Int64Type>::from_iter_values(0..128); |
| 1321 | let slice1 = arr.slice(0, 64); |
| 1322 | let slice2 = arr.slice(64, 64); |
| 1323 | |
| 1324 | // both slices report the full buffer memory usage, even though the buffers are shared |
| 1325 | assert_eq!(slice1.get_array_memory_size(), arr.get_array_memory_size()); |
| 1326 | assert_eq!(slice2.get_array_memory_size(), arr.get_array_memory_size()); |
| 1327 | } |
| 1328 | |
| 1329 | #[test] |
| 1330 | fn test_memory_size_primitive_nullable() { |
| 1331 | let arr: PrimitiveArray<Int64Type> = (0..128) |
| 1332 | .map(|i| if i % 20 == 0 { Some(i) } else { None }) |
| 1333 | .collect(); |
| 1334 | let empty_with_bitmap = PrimitiveArray::<Int64Type>::from( |
| 1335 | ArrayData::builder(arr.data_type().clone()) |
| 1336 | .add_buffer(MutableBuffer::new(0).into()) |
| 1337 | .null_bit_buffer(Some(MutableBuffer::new_null(0).into())) |
| 1338 | .build() |
| 1339 | .unwrap(), |
| 1340 | ); |
| 1341 | |
| 1342 | // expected size is the size of the PrimitiveArray struct, |
| 1343 | // which includes the optional validity buffer |
| 1344 | // plus one buffer on the heap |
| 1345 | assert_eq!( |
| 1346 | std::mem::size_of::<PrimitiveArray<Int64Type>>(), |
| 1347 | empty_with_bitmap.get_array_memory_size() |
| 1348 | ); |
| 1349 | |
| 1350 | // subtract empty array to avoid magic numbers for the size of additional fields |
| 1351 | // the size of the validity bitmap is rounded up to 64 bytes |
| 1352 | assert_eq!( |
| 1353 | arr.get_array_memory_size() - empty_with_bitmap.get_array_memory_size(), |
| 1354 | 128 * std::mem::size_of::<i64>() + 64 |
| 1355 | ); |
| 1356 | } |
| 1357 | |
| 1358 | #[test] |
| 1359 | fn test_memory_size_dictionary() { |
| 1360 | let values = PrimitiveArray::<Int64Type>::from_iter_values(0..16); |
| 1361 | let keys = PrimitiveArray::<Int16Type>::from_iter_values( |
| 1362 | (0..256).map(|i| (i % values.len()) as i16), |
| 1363 | ); |
| 1364 | |
| 1365 | let dict_data_type = DataType::Dictionary( |
| 1366 | Box::new(keys.data_type().clone()), |
| 1367 | Box::new(values.data_type().clone()), |
| 1368 | ); |
| 1369 | let dict_data = keys |
| 1370 | .into_data() |
| 1371 | .into_builder() |
| 1372 | .data_type(dict_data_type) |
| 1373 | .child_data(vec![values.into_data()]) |
| 1374 | .build() |
| 1375 | .unwrap(); |
| 1376 | |
| 1377 | let empty_data = ArrayData::new_empty(&DataType::Dictionary( |
| 1378 | Box::new(DataType::Int16), |
| 1379 | Box::new(DataType::Int64), |
| 1380 | )); |
| 1381 | |
| 1382 | let arr = DictionaryArray::<Int16Type>::from(dict_data); |
| 1383 | let empty = DictionaryArray::<Int16Type>::from(empty_data); |
| 1384 | |
| 1385 | let expected_keys_size = 256 * std::mem::size_of::<i16>(); |
| 1386 | assert_eq!( |
| 1387 | arr.keys().get_array_memory_size() - empty.keys().get_array_memory_size(), |
| 1388 | expected_keys_size |
| 1389 | ); |
| 1390 | |
| 1391 | let expected_values_size = 16 * std::mem::size_of::<i64>(); |
| 1392 | assert_eq!( |
| 1393 | arr.values().get_array_memory_size() - empty.values().get_array_memory_size(), |
| 1394 | expected_values_size |
| 1395 | ); |
| 1396 | |
| 1397 | let expected_size = expected_keys_size + expected_values_size; |
| 1398 | assert_eq!( |
| 1399 | arr.get_array_memory_size() - empty.get_array_memory_size(), |
| 1400 | expected_size |
| 1401 | ); |
| 1402 | } |
| 1403 | |
| 1404 | /// Test function that takes an &dyn Array |
| 1405 | fn compute_my_thing(arr: &dyn Array) -> bool { |
| 1406 | !arr.is_empty() |
| 1407 | } |
| 1408 | |
| 1409 | #[test] |
| 1410 | fn test_array_ref_as_array() { |
| 1411 | let arr: Int32Array = vec![1, 2, 3].into_iter().map(Some).collect(); |
| 1412 | |
| 1413 | // works well! |
| 1414 | assert!(compute_my_thing(&arr)); |
| 1415 | |
| 1416 | // Should also work when wrapped as an ArrayRef |
| 1417 | let arr: ArrayRef = Arc::new(arr); |
| 1418 | assert!(compute_my_thing(&arr)); |
| 1419 | assert!(compute_my_thing(arr.as_ref())); |
| 1420 | } |
| 1421 | |
| 1422 | #[test] |
| 1423 | fn test_downcast_array() { |
| 1424 | let array: Int32Array = vec![1, 2, 3].into_iter().map(Some).collect(); |
| 1425 | |
| 1426 | let boxed: ArrayRef = Arc::new(array); |
| 1427 | let array: Int32Array = downcast_array(&boxed); |
| 1428 | |
| 1429 | let expected: Int32Array = vec![1, 2, 3].into_iter().map(Some).collect(); |
| 1430 | assert_eq!(array, expected); |
| 1431 | } |
| 1432 | } |