001/* 002 * Copyright (C) 2015 The Guava Authors 003 * 004 * Licensed under the Apache License, Version 2.0 (the "License"); you 005 * may not use this file except in compliance with the License. You may 006 * obtain a copy of the License at 007 * 008 * http://www.apache.org/licenses/LICENSE-2.0 009 * 010 * Unless required by applicable law or agreed to in writing, software 011 * distributed under the License is distributed on an "AS IS" BASIS, 012 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 013 * implied. See the License for the specific language governing 014 * permissions and limitations under the License. 015 */ 016 017package com.google.common.collect; 018 019import static com.google.common.base.Preconditions.checkNotNull; 020import static com.google.common.base.Preconditions.checkState; 021 022import com.google.common.annotations.Beta; 023import com.google.common.annotations.GwtCompatible; 024import com.google.common.math.LongMath; 025import java.util.ArrayDeque; 026import java.util.Collection; 027import java.util.Deque; 028import java.util.Iterator; 029import java.util.OptionalDouble; 030import java.util.OptionalInt; 031import java.util.OptionalLong; 032import java.util.PrimitiveIterator; 033import java.util.Spliterator; 034import java.util.Spliterators; 035import java.util.Spliterators.AbstractSpliterator; 036import java.util.function.BiConsumer; 037import java.util.function.BiFunction; 038import java.util.function.Consumer; 039import java.util.function.DoubleConsumer; 040import java.util.function.IntConsumer; 041import java.util.function.LongConsumer; 042import java.util.stream.DoubleStream; 043import java.util.stream.IntStream; 044import java.util.stream.LongStream; 045import java.util.stream.Stream; 046import java.util.stream.StreamSupport; 047import org.checkerframework.checker.nullness.qual.Nullable; 048 049/** 050 * Static utility methods related to {@code Stream} instances. 051 * 052 * @since 21.0 053 */ 054@GwtCompatible 055public final class Streams { 056 /** 057 * Returns a sequential {@link Stream} of the contents of {@code iterable}, delegating to {@link 058 * Collection#stream} if possible. 059 */ 060 public static <T> Stream<T> stream(Iterable<T> iterable) { 061 return (iterable instanceof Collection) 062 ? ((Collection<T>) iterable).stream() 063 : StreamSupport.stream(iterable.spliterator(), false); 064 } 065 066 /** 067 * Returns {@link Collection#stream}. 068 * 069 * @deprecated There is no reason to use this; just invoke {@code collection.stream()} directly. 070 */ 071 @Beta 072 @Deprecated 073 public static <T> Stream<T> stream(Collection<T> collection) { 074 return collection.stream(); 075 } 076 077 /** 078 * Returns a sequential {@link Stream} of the remaining contents of {@code iterator}. Do not use 079 * {@code iterator} directly after passing it to this method. 080 */ 081 @Beta 082 public static <T> Stream<T> stream(Iterator<T> iterator) { 083 return StreamSupport.stream(Spliterators.spliteratorUnknownSize(iterator, 0), false); 084 } 085 086 /** 087 * If a value is present in {@code optional}, returns a stream containing only that element, 088 * otherwise returns an empty stream. 089 */ 090 @Beta 091 public static <T> Stream<T> stream(com.google.common.base.Optional<T> optional) { 092 return optional.isPresent() ? Stream.of(optional.get()) : Stream.empty(); 093 } 094 095 /** 096 * If a value is present in {@code optional}, returns a stream containing only that element, 097 * otherwise returns an empty stream. 098 * 099 * <p><b>Java 9 users:</b> use {@code optional.stream()} instead. 100 */ 101 @Beta 102 public static <T> Stream<T> stream(java.util.Optional<T> optional) { 103 return optional.isPresent() ? Stream.of(optional.get()) : Stream.empty(); 104 } 105 106 /** 107 * If a value is present in {@code optional}, returns a stream containing only that element, 108 * otherwise returns an empty stream. 109 * 110 * <p><b>Java 9 users:</b> use {@code optional.stream()} instead. 111 */ 112 @Beta 113 public static IntStream stream(OptionalInt optional) { 114 return optional.isPresent() ? IntStream.of(optional.getAsInt()) : IntStream.empty(); 115 } 116 117 /** 118 * If a value is present in {@code optional}, returns a stream containing only that element, 119 * otherwise returns an empty stream. 120 * 121 * <p><b>Java 9 users:</b> use {@code optional.stream()} instead. 122 */ 123 @Beta 124 public static LongStream stream(OptionalLong optional) { 125 return optional.isPresent() ? LongStream.of(optional.getAsLong()) : LongStream.empty(); 126 } 127 128 /** 129 * If a value is present in {@code optional}, returns a stream containing only that element, 130 * otherwise returns an empty stream. 131 * 132 * <p><b>Java 9 users:</b> use {@code optional.stream()} instead. 133 */ 134 @Beta 135 public static DoubleStream stream(OptionalDouble optional) { 136 return optional.isPresent() ? DoubleStream.of(optional.getAsDouble()) : DoubleStream.empty(); 137 } 138 139 /** 140 * Returns a {@link Stream} containing the elements of the first stream, followed by the elements 141 * of the second stream, and so on. 142 * 143 * <p>This is equivalent to {@code Stream.of(streams).flatMap(stream -> stream)}, but the returned 144 * stream may perform better. 145 * 146 * @see Stream#concat(Stream, Stream) 147 */ 148 @SafeVarargs 149 public static <T> Stream<T> concat(Stream<? extends T>... streams) { 150 // TODO(lowasser): consider an implementation that can support SUBSIZED 151 boolean isParallel = false; 152 int characteristics = Spliterator.ORDERED | Spliterator.SIZED | Spliterator.NONNULL; 153 long estimatedSize = 0L; 154 ImmutableList.Builder<Spliterator<? extends T>> splitrsBuilder = 155 new ImmutableList.Builder<>(streams.length); 156 for (Stream<? extends T> stream : streams) { 157 isParallel |= stream.isParallel(); 158 Spliterator<? extends T> splitr = stream.spliterator(); 159 splitrsBuilder.add(splitr); 160 characteristics &= splitr.characteristics(); 161 estimatedSize = LongMath.saturatedAdd(estimatedSize, splitr.estimateSize()); 162 } 163 return StreamSupport.stream( 164 CollectSpliterators.flatMap( 165 splitrsBuilder.build().spliterator(), 166 splitr -> (Spliterator<T>) splitr, 167 characteristics, 168 estimatedSize), 169 isParallel) 170 .onClose( 171 () -> { 172 for (Stream<? extends T> stream : streams) { 173 stream.close(); 174 } 175 }); 176 } 177 178 /** 179 * Returns an {@link IntStream} containing the elements of the first stream, followed by the 180 * elements of the second stream, and so on. 181 * 182 * <p>This is equivalent to {@code Stream.of(streams).flatMapToInt(stream -> stream)}, but the 183 * returned stream may perform better. 184 * 185 * @see IntStream#concat(IntStream, IntStream) 186 */ 187 public static IntStream concat(IntStream... streams) { 188 // TODO(lowasser): optimize this later 189 return Stream.of(streams).flatMapToInt(stream -> stream); 190 } 191 192 /** 193 * Returns a {@link LongStream} containing the elements of the first stream, followed by the 194 * elements of the second stream, and so on. 195 * 196 * <p>This is equivalent to {@code Stream.of(streams).flatMapToLong(stream -> stream)}, but the 197 * returned stream may perform better. 198 * 199 * @see LongStream#concat(LongStream, LongStream) 200 */ 201 public static LongStream concat(LongStream... streams) { 202 // TODO(lowasser): optimize this later 203 return Stream.of(streams).flatMapToLong(stream -> stream); 204 } 205 206 /** 207 * Returns a {@link DoubleStream} containing the elements of the first stream, followed by the 208 * elements of the second stream, and so on. 209 * 210 * <p>This is equivalent to {@code Stream.of(streams).flatMapToDouble(stream -> stream)}, but the 211 * returned stream may perform better. 212 * 213 * @see DoubleStream#concat(DoubleStream, DoubleStream) 214 */ 215 public static DoubleStream concat(DoubleStream... streams) { 216 // TODO(lowasser): optimize this later 217 return Stream.of(streams).flatMapToDouble(stream -> stream); 218 } 219 220 /** 221 * Returns a stream in which each element is the result of passing the corresponding element of 222 * each of {@code streamA} and {@code streamB} to {@code function}. 223 * 224 * <p>For example: 225 * 226 * <pre>{@code 227 * Streams.zip( 228 * Stream.of("foo1", "foo2", "foo3"), 229 * Stream.of("bar1", "bar2"), 230 * (arg1, arg2) -> arg1 + ":" + arg2) 231 * }</pre> 232 * 233 * <p>will return {@code Stream.of("foo1:bar1", "foo2:bar2")}. 234 * 235 * <p>The resulting stream will only be as long as the shorter of the two input streams; if one 236 * stream is longer, its extra elements will be ignored. 237 * 238 * <p>Note that if you are calling {@link Stream#forEach} on the resulting stream, you might want 239 * to consider using {@link #forEachPair} instead of this method. 240 * 241 * <p><b>Performance note:</b> The resulting stream is not <a 242 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a>. 243 * This may harm parallel performance. 244 */ 245 @Beta 246 public static <A, B, R> Stream<R> zip( 247 Stream<A> streamA, Stream<B> streamB, BiFunction<? super A, ? super B, R> function) { 248 checkNotNull(streamA); 249 checkNotNull(streamB); 250 checkNotNull(function); 251 boolean isParallel = streamA.isParallel() || streamB.isParallel(); // same as Stream.concat 252 Spliterator<A> splitrA = streamA.spliterator(); 253 Spliterator<B> splitrB = streamB.spliterator(); 254 int characteristics = 255 splitrA.characteristics() 256 & splitrB.characteristics() 257 & (Spliterator.SIZED | Spliterator.ORDERED); 258 Iterator<A> itrA = Spliterators.iterator(splitrA); 259 Iterator<B> itrB = Spliterators.iterator(splitrB); 260 return StreamSupport.stream( 261 new AbstractSpliterator<R>( 262 Math.min(splitrA.estimateSize(), splitrB.estimateSize()), characteristics) { 263 @Override 264 public boolean tryAdvance(Consumer<? super R> action) { 265 if (itrA.hasNext() && itrB.hasNext()) { 266 action.accept(function.apply(itrA.next(), itrB.next())); 267 return true; 268 } 269 return false; 270 } 271 }, 272 isParallel) 273 .onClose(streamA::close) 274 .onClose(streamB::close); 275 } 276 277 /** 278 * Invokes {@code consumer} once for each pair of <i>corresponding</i> elements in {@code streamA} 279 * and {@code streamB}. If one stream is longer than the other, the extra elements are silently 280 * ignored. Elements passed to the consumer are guaranteed to come from the same position in their 281 * respective source streams. For example: 282 * 283 * <pre>{@code 284 * Streams.forEachPair( 285 * Stream.of("foo1", "foo2", "foo3"), 286 * Stream.of("bar1", "bar2"), 287 * (arg1, arg2) -> System.out.println(arg1 + ":" + arg2) 288 * }</pre> 289 * 290 * <p>will print: 291 * 292 * <pre>{@code 293 * foo1:bar1 294 * foo2:bar2 295 * }</pre> 296 * 297 * <p><b>Warning:</b> If either supplied stream is a parallel stream, the same correspondence 298 * between elements will be made, but the order in which those pairs of elements are passed to the 299 * consumer is <i>not</i> defined. 300 * 301 * <p>Note that many usages of this method can be replaced with simpler calls to {@link #zip}. 302 * This method behaves equivalently to {@linkplain #zip zipping} the stream elements into 303 * temporary pair objects and then using {@link Stream#forEach} on that stream. 304 * 305 * @since 22.0 306 */ 307 @Beta 308 public static <A, B> void forEachPair( 309 Stream<A> streamA, Stream<B> streamB, BiConsumer<? super A, ? super B> consumer) { 310 checkNotNull(consumer); 311 312 if (streamA.isParallel() || streamB.isParallel()) { 313 zip(streamA, streamB, TemporaryPair::new).forEach(pair -> consumer.accept(pair.a, pair.b)); 314 } else { 315 Iterator<A> iterA = streamA.iterator(); 316 Iterator<B> iterB = streamB.iterator(); 317 while (iterA.hasNext() && iterB.hasNext()) { 318 consumer.accept(iterA.next(), iterB.next()); 319 } 320 } 321 } 322 323 // Use this carefully - it doesn't implement value semantics 324 private static class TemporaryPair<A, B> { 325 final A a; 326 final B b; 327 328 TemporaryPair(A a, B b) { 329 this.a = a; 330 this.b = b; 331 } 332 } 333 334 /** 335 * Returns a stream consisting of the results of applying the given function to the elements of 336 * {@code stream} and their indices in the stream. For example, 337 * 338 * <pre>{@code 339 * mapWithIndex( 340 * Stream.of("a", "b", "c"), 341 * (str, index) -> str + ":" + index) 342 * }</pre> 343 * 344 * <p>would return {@code Stream.of("a:0", "b:1", "c:2")}. 345 * 346 * <p>The resulting stream is <a 347 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a> 348 * if and only if {@code stream} was efficiently splittable and its underlying spliterator 349 * reported {@link Spliterator#SUBSIZED}. This is generally the case if the underlying stream 350 * comes from a data structure supporting efficient indexed random access, typically an array or 351 * list. 352 * 353 * <p>The order of the resulting stream is defined if and only if the order of the original stream 354 * was defined. 355 */ 356 @Beta 357 public static <T, R> Stream<R> mapWithIndex( 358 Stream<T> stream, FunctionWithIndex<? super T, ? extends R> function) { 359 checkNotNull(stream); 360 checkNotNull(function); 361 boolean isParallel = stream.isParallel(); 362 Spliterator<T> fromSpliterator = stream.spliterator(); 363 364 if (!fromSpliterator.hasCharacteristics(Spliterator.SUBSIZED)) { 365 Iterator<T> fromIterator = Spliterators.iterator(fromSpliterator); 366 return StreamSupport.stream( 367 new AbstractSpliterator<R>( 368 fromSpliterator.estimateSize(), 369 fromSpliterator.characteristics() & (Spliterator.ORDERED | Spliterator.SIZED)) { 370 long index = 0; 371 372 @Override 373 public boolean tryAdvance(Consumer<? super R> action) { 374 if (fromIterator.hasNext()) { 375 action.accept(function.apply(fromIterator.next(), index++)); 376 return true; 377 } 378 return false; 379 } 380 }, 381 isParallel) 382 .onClose(stream::close); 383 } 384 class Splitr extends MapWithIndexSpliterator<Spliterator<T>, R, Splitr> implements Consumer<T> { 385 @Nullable T holder; 386 387 Splitr(Spliterator<T> splitr, long index) { 388 super(splitr, index); 389 } 390 391 @Override 392 public void accept(@Nullable T t) { 393 this.holder = t; 394 } 395 396 @Override 397 public boolean tryAdvance(Consumer<? super R> action) { 398 if (fromSpliterator.tryAdvance(this)) { 399 try { 400 action.accept(function.apply(holder, index++)); 401 return true; 402 } finally { 403 holder = null; 404 } 405 } 406 return false; 407 } 408 409 @Override 410 Splitr createSplit(Spliterator<T> from, long i) { 411 return new Splitr(from, i); 412 } 413 } 414 return StreamSupport.stream(new Splitr(fromSpliterator, 0), isParallel).onClose(stream::close); 415 } 416 417 /** 418 * Returns a stream consisting of the results of applying the given function to the elements of 419 * {@code stream} and their indexes in the stream. For example, 420 * 421 * <pre>{@code 422 * mapWithIndex( 423 * IntStream.of(0, 1, 2), 424 * (i, index) -> i + ":" + index) 425 * }</pre> 426 * 427 * <p>...would return {@code Stream.of("0:0", "1:1", "2:2")}. 428 * 429 * <p>The resulting stream is <a 430 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a> 431 * if and only if {@code stream} was efficiently splittable and its underlying spliterator 432 * reported {@link Spliterator#SUBSIZED}. This is generally the case if the underlying stream 433 * comes from a data structure supporting efficient indexed random access, typically an array or 434 * list. 435 * 436 * <p>The order of the resulting stream is defined if and only if the order of the original stream 437 * was defined. 438 */ 439 @Beta 440 public static <R> Stream<R> mapWithIndex(IntStream stream, IntFunctionWithIndex<R> function) { 441 checkNotNull(stream); 442 checkNotNull(function); 443 boolean isParallel = stream.isParallel(); 444 Spliterator.OfInt fromSpliterator = stream.spliterator(); 445 446 if (!fromSpliterator.hasCharacteristics(Spliterator.SUBSIZED)) { 447 PrimitiveIterator.OfInt fromIterator = Spliterators.iterator(fromSpliterator); 448 return StreamSupport.stream( 449 new AbstractSpliterator<R>( 450 fromSpliterator.estimateSize(), 451 fromSpliterator.characteristics() & (Spliterator.ORDERED | Spliterator.SIZED)) { 452 long index = 0; 453 454 @Override 455 public boolean tryAdvance(Consumer<? super R> action) { 456 if (fromIterator.hasNext()) { 457 action.accept(function.apply(fromIterator.nextInt(), index++)); 458 return true; 459 } 460 return false; 461 } 462 }, 463 isParallel) 464 .onClose(stream::close); 465 } 466 class Splitr extends MapWithIndexSpliterator<Spliterator.OfInt, R, Splitr> 467 implements IntConsumer, Spliterator<R> { 468 int holder; 469 470 Splitr(Spliterator.OfInt splitr, long index) { 471 super(splitr, index); 472 } 473 474 @Override 475 public void accept(int t) { 476 this.holder = t; 477 } 478 479 @Override 480 public boolean tryAdvance(Consumer<? super R> action) { 481 if (fromSpliterator.tryAdvance(this)) { 482 action.accept(function.apply(holder, index++)); 483 return true; 484 } 485 return false; 486 } 487 488 @Override 489 Splitr createSplit(Spliterator.OfInt from, long i) { 490 return new Splitr(from, i); 491 } 492 } 493 return StreamSupport.stream(new Splitr(fromSpliterator, 0), isParallel).onClose(stream::close); 494 } 495 496 /** 497 * Returns a stream consisting of the results of applying the given function to the elements of 498 * {@code stream} and their indexes in the stream. For example, 499 * 500 * <pre>{@code 501 * mapWithIndex( 502 * LongStream.of(0, 1, 2), 503 * (i, index) -> i + ":" + index) 504 * }</pre> 505 * 506 * <p>...would return {@code Stream.of("0:0", "1:1", "2:2")}. 507 * 508 * <p>The resulting stream is <a 509 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a> 510 * if and only if {@code stream} was efficiently splittable and its underlying spliterator 511 * reported {@link Spliterator#SUBSIZED}. This is generally the case if the underlying stream 512 * comes from a data structure supporting efficient indexed random access, typically an array or 513 * list. 514 * 515 * <p>The order of the resulting stream is defined if and only if the order of the original stream 516 * was defined. 517 */ 518 @Beta 519 public static <R> Stream<R> mapWithIndex(LongStream stream, LongFunctionWithIndex<R> function) { 520 checkNotNull(stream); 521 checkNotNull(function); 522 boolean isParallel = stream.isParallel(); 523 Spliterator.OfLong fromSpliterator = stream.spliterator(); 524 525 if (!fromSpliterator.hasCharacteristics(Spliterator.SUBSIZED)) { 526 PrimitiveIterator.OfLong fromIterator = Spliterators.iterator(fromSpliterator); 527 return StreamSupport.stream( 528 new AbstractSpliterator<R>( 529 fromSpliterator.estimateSize(), 530 fromSpliterator.characteristics() & (Spliterator.ORDERED | Spliterator.SIZED)) { 531 long index = 0; 532 533 @Override 534 public boolean tryAdvance(Consumer<? super R> action) { 535 if (fromIterator.hasNext()) { 536 action.accept(function.apply(fromIterator.nextLong(), index++)); 537 return true; 538 } 539 return false; 540 } 541 }, 542 isParallel) 543 .onClose(stream::close); 544 } 545 class Splitr extends MapWithIndexSpliterator<Spliterator.OfLong, R, Splitr> 546 implements LongConsumer, Spliterator<R> { 547 long holder; 548 549 Splitr(Spliterator.OfLong splitr, long index) { 550 super(splitr, index); 551 } 552 553 @Override 554 public void accept(long t) { 555 this.holder = t; 556 } 557 558 @Override 559 public boolean tryAdvance(Consumer<? super R> action) { 560 if (fromSpliterator.tryAdvance(this)) { 561 action.accept(function.apply(holder, index++)); 562 return true; 563 } 564 return false; 565 } 566 567 @Override 568 Splitr createSplit(Spliterator.OfLong from, long i) { 569 return new Splitr(from, i); 570 } 571 } 572 return StreamSupport.stream(new Splitr(fromSpliterator, 0), isParallel).onClose(stream::close); 573 } 574 575 /** 576 * Returns a stream consisting of the results of applying the given function to the elements of 577 * {@code stream} and their indexes in the stream. For example, 578 * 579 * <pre>{@code 580 * mapWithIndex( 581 * DoubleStream.of(0, 1, 2), 582 * (x, index) -> x + ":" + index) 583 * }</pre> 584 * 585 * <p>...would return {@code Stream.of("0.0:0", "1.0:1", "2.0:2")}. 586 * 587 * <p>The resulting stream is <a 588 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a> 589 * if and only if {@code stream} was efficiently splittable and its underlying spliterator 590 * reported {@link Spliterator#SUBSIZED}. This is generally the case if the underlying stream 591 * comes from a data structure supporting efficient indexed random access, typically an array or 592 * list. 593 * 594 * <p>The order of the resulting stream is defined if and only if the order of the original stream 595 * was defined. 596 */ 597 @Beta 598 public static <R> Stream<R> mapWithIndex( 599 DoubleStream stream, DoubleFunctionWithIndex<R> function) { 600 checkNotNull(stream); 601 checkNotNull(function); 602 boolean isParallel = stream.isParallel(); 603 Spliterator.OfDouble fromSpliterator = stream.spliterator(); 604 605 if (!fromSpliterator.hasCharacteristics(Spliterator.SUBSIZED)) { 606 PrimitiveIterator.OfDouble fromIterator = Spliterators.iterator(fromSpliterator); 607 return StreamSupport.stream( 608 new AbstractSpliterator<R>( 609 fromSpliterator.estimateSize(), 610 fromSpliterator.characteristics() & (Spliterator.ORDERED | Spliterator.SIZED)) { 611 long index = 0; 612 613 @Override 614 public boolean tryAdvance(Consumer<? super R> action) { 615 if (fromIterator.hasNext()) { 616 action.accept(function.apply(fromIterator.nextDouble(), index++)); 617 return true; 618 } 619 return false; 620 } 621 }, 622 isParallel) 623 .onClose(stream::close); 624 } 625 class Splitr extends MapWithIndexSpliterator<Spliterator.OfDouble, R, Splitr> 626 implements DoubleConsumer, Spliterator<R> { 627 double holder; 628 629 Splitr(Spliterator.OfDouble splitr, long index) { 630 super(splitr, index); 631 } 632 633 @Override 634 public void accept(double t) { 635 this.holder = t; 636 } 637 638 @Override 639 public boolean tryAdvance(Consumer<? super R> action) { 640 if (fromSpliterator.tryAdvance(this)) { 641 action.accept(function.apply(holder, index++)); 642 return true; 643 } 644 return false; 645 } 646 647 @Override 648 Splitr createSplit(Spliterator.OfDouble from, long i) { 649 return new Splitr(from, i); 650 } 651 } 652 return StreamSupport.stream(new Splitr(fromSpliterator, 0), isParallel).onClose(stream::close); 653 } 654 655 /** 656 * An analogue of {@link java.util.function.Function} also accepting an index. 657 * 658 * <p>This interface is only intended for use by callers of {@link #mapWithIndex(Stream, 659 * FunctionWithIndex)}. 660 * 661 * @since 21.0 662 */ 663 @Beta 664 public interface FunctionWithIndex<T, R> { 665 /** Applies this function to the given argument and its index within a stream. */ 666 R apply(T from, long index); 667 } 668 669 private abstract static class MapWithIndexSpliterator< 670 F extends Spliterator<?>, R, S extends MapWithIndexSpliterator<F, R, S>> 671 implements Spliterator<R> { 672 final F fromSpliterator; 673 long index; 674 675 MapWithIndexSpliterator(F fromSpliterator, long index) { 676 this.fromSpliterator = fromSpliterator; 677 this.index = index; 678 } 679 680 abstract S createSplit(F from, long i); 681 682 @Override 683 public S trySplit() { 684 @SuppressWarnings("unchecked") 685 F split = (F) fromSpliterator.trySplit(); 686 if (split == null) { 687 return null; 688 } 689 S result = createSplit(split, index); 690 this.index += split.getExactSizeIfKnown(); 691 return result; 692 } 693 694 @Override 695 public long estimateSize() { 696 return fromSpliterator.estimateSize(); 697 } 698 699 @Override 700 public int characteristics() { 701 return fromSpliterator.characteristics() 702 & (Spliterator.ORDERED | Spliterator.SIZED | Spliterator.SUBSIZED); 703 } 704 } 705 706 /** 707 * An analogue of {@link java.util.function.IntFunction} also accepting an index. 708 * 709 * <p>This interface is only intended for use by callers of {@link #mapWithIndex(IntStream, 710 * IntFunctionWithIndex)}. 711 * 712 * @since 21.0 713 */ 714 @Beta 715 public interface IntFunctionWithIndex<R> { 716 /** Applies this function to the given argument and its index within a stream. */ 717 R apply(int from, long index); 718 } 719 720 /** 721 * An analogue of {@link java.util.function.LongFunction} also accepting an index. 722 * 723 * <p>This interface is only intended for use by callers of {@link #mapWithIndex(LongStream, 724 * LongFunctionWithIndex)}. 725 * 726 * @since 21.0 727 */ 728 @Beta 729 public interface LongFunctionWithIndex<R> { 730 /** Applies this function to the given argument and its index within a stream. */ 731 R apply(long from, long index); 732 } 733 734 /** 735 * An analogue of {@link java.util.function.DoubleFunction} also accepting an index. 736 * 737 * <p>This interface is only intended for use by callers of {@link #mapWithIndex(DoubleStream, 738 * DoubleFunctionWithIndex)}. 739 * 740 * @since 21.0 741 */ 742 @Beta 743 public interface DoubleFunctionWithIndex<R> { 744 /** Applies this function to the given argument and its index within a stream. */ 745 R apply(double from, long index); 746 } 747 748 /** 749 * Returns the last element of the specified stream, or {@link java.util.Optional#empty} if the 750 * stream is empty. 751 * 752 * <p>Equivalent to {@code stream.reduce((a, b) -> b)}, but may perform significantly better. This 753 * method's runtime will be between O(log n) and O(n), performing better on <a 754 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a> 755 * streams. 756 * 757 * <p>If the stream has nondeterministic order, this has equivalent semantics to {@link 758 * Stream#findAny} (which you might as well use). 759 * 760 * @see Stream#findFirst() 761 * @throws NullPointerException if the last element of the stream is null 762 */ 763 @Beta 764 public static <T> java.util.Optional<T> findLast(Stream<T> stream) { 765 class OptionalState { 766 boolean set = false; 767 T value = null; 768 769 void set(@Nullable T value) { 770 this.set = true; 771 this.value = value; 772 } 773 774 T get() { 775 checkState(set); 776 return value; 777 } 778 } 779 OptionalState state = new OptionalState(); 780 781 Deque<Spliterator<T>> splits = new ArrayDeque<>(); 782 splits.addLast(stream.spliterator()); 783 784 while (!splits.isEmpty()) { 785 Spliterator<T> spliterator = splits.removeLast(); 786 787 if (spliterator.getExactSizeIfKnown() == 0) { 788 continue; // drop this split 789 } 790 791 // Many spliterators will have trySplits that are SUBSIZED even if they are not themselves 792 // SUBSIZED. 793 if (spliterator.hasCharacteristics(Spliterator.SUBSIZED)) { 794 // we can drill down to exactly the smallest nonempty spliterator 795 while (true) { 796 Spliterator<T> prefix = spliterator.trySplit(); 797 if (prefix == null || prefix.getExactSizeIfKnown() == 0) { 798 break; 799 } else if (spliterator.getExactSizeIfKnown() == 0) { 800 spliterator = prefix; 801 break; 802 } 803 } 804 805 // spliterator is known to be nonempty now 806 spliterator.forEachRemaining(state::set); 807 return java.util.Optional.of(state.get()); 808 } 809 810 Spliterator<T> prefix = spliterator.trySplit(); 811 if (prefix == null || prefix.getExactSizeIfKnown() == 0) { 812 // we can't split this any further 813 spliterator.forEachRemaining(state::set); 814 if (state.set) { 815 return java.util.Optional.of(state.get()); 816 } 817 // fall back to the last split 818 continue; 819 } 820 splits.addLast(prefix); 821 splits.addLast(spliterator); 822 } 823 return java.util.Optional.empty(); 824 } 825 826 /** 827 * Returns the last element of the specified stream, or {@link OptionalInt#empty} if the stream is 828 * empty. 829 * 830 * <p>Equivalent to {@code stream.reduce((a, b) -> b)}, but may perform significantly better. This 831 * method's runtime will be between O(log n) and O(n), performing better on <a 832 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a> 833 * streams. 834 * 835 * @see IntStream#findFirst() 836 * @throws NullPointerException if the last element of the stream is null 837 */ 838 @Beta 839 public static OptionalInt findLast(IntStream stream) { 840 // findLast(Stream) does some allocation, so we might as well box some more 841 java.util.Optional<Integer> boxedLast = findLast(stream.boxed()); 842 return boxedLast.isPresent() ? OptionalInt.of(boxedLast.get()) : OptionalInt.empty(); 843 } 844 845 /** 846 * Returns the last element of the specified stream, or {@link OptionalLong#empty} if the stream 847 * is empty. 848 * 849 * <p>Equivalent to {@code stream.reduce((a, b) -> b)}, but may perform significantly better. This 850 * method's runtime will be between O(log n) and O(n), performing better on <a 851 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a> 852 * streams. 853 * 854 * @see LongStream#findFirst() 855 * @throws NullPointerException if the last element of the stream is null 856 */ 857 @Beta 858 public static OptionalLong findLast(LongStream stream) { 859 // findLast(Stream) does some allocation, so we might as well box some more 860 java.util.Optional<Long> boxedLast = findLast(stream.boxed()); 861 return boxedLast.isPresent() ? OptionalLong.of(boxedLast.get()) : OptionalLong.empty(); 862 } 863 864 /** 865 * Returns the last element of the specified stream, or {@link OptionalDouble#empty} if the stream 866 * is empty. 867 * 868 * <p>Equivalent to {@code stream.reduce((a, b) -> b)}, but may perform significantly better. This 869 * method's runtime will be between O(log n) and O(n), performing better on <a 870 * href="http://gee.cs.oswego.edu/dl/html/StreamParallelGuidance.html">efficiently splittable</a> 871 * streams. 872 * 873 * @see DoubleStream#findFirst() 874 * @throws NullPointerException if the last element of the stream is null 875 */ 876 @Beta 877 public static OptionalDouble findLast(DoubleStream stream) { 878 // findLast(Stream) does some allocation, so we might as well box some more 879 java.util.Optional<Double> boxedLast = findLast(stream.boxed()); 880 return boxedLast.isPresent() ? OptionalDouble.of(boxedLast.get()) : OptionalDouble.empty(); 881 } 882 883 private Streams() {} 884}