Module iter

Expand description

Traits for writing parallel programs using an iterator-style interface

You will rarely need to interact with this module directly unless you have need to name one of the iterator types.

Parallel iterators make it easy to write iterator-like chains that execute in parallel: typically all you have to do is convert the first .iter() (or iter_mut(), into_iter(), etc) method into par_iter() (or par_iter_mut(), into_par_iter(), etc). For example, to compute the sum of the squares of a sequence of integers, one might write:

use rayon::prelude::*;
fn sum_of_squares(input: &[i32]) -> i32 {
    input.par_iter()
         .map(|i| i * i)
         .sum()
}

Or, to increment all the integers in a slice, you could write:

use rayon::prelude::*;
fn increment_all(input: &mut [i32]) {
    input.par_iter_mut()
         .for_each(|p| *p += 1);
}

To use parallel iterators, first import the traits by adding something like use rayon::prelude::* to your module. You can then call par_iter, par_iter_mut, or into_par_iter to get a parallel iterator. Like a regular iterator, parallel iterators work by first constructing a computation and then executing it.

In addition to par_iter() and friends, some types offer other ways to create (or consume) parallel iterators:

Slices (&[T], &mut [T]) offer methods like par_split and par_windows, as well as various parallel sorting operations. See the ParallelSlice trait for the full list.
Strings (&str) offer methods like par_split and par_lines. See the ParallelString trait for the full list.
Various collections offer par_extend, which grows a collection given a parallel iterator. (If you don’t have a collection to extend, you can use collect() to create a new one from scratch.)

To see the full range of methods available on parallel iterators, check out the ParallelIterator and IndexedParallelIterator traits.

If you’d like to build a custom parallel iterator, or to write your own combinator, then check out the split function and the plumbing module.

Note: Several of the ParallelIterator methods rely on a Try trait which has been deliberately obscured from the public API. This trait is intended to mirror the unstable std::ops::Try with implementations for Option and Result, where Some/Ok values will let those iterators continue, but None/Err values will exit early.

A note about object safety: It is currently not possible to wrap a ParallelIterator (or any trait that depends on it) using a Box<dyn ParallelIterator> or other kind of dynamic allocation, because ParallelIterator is not object-safe. (This keeps the implementation simpler and allows extra optimizations.)

Modules§

plumbing: Traits and functions used to implement parallel iteration. These are low-level details – users of parallel iterators should not need to interact with them directly. See the plumbing README for a general overview.

Structs§

Chain: Chain is an iterator that joins b after a in one continuous iterator. This struct is created by the chain() method on ParallelIterator
Chunks: Chunks is an iterator that groups elements of an underlying iterator.
Cloned: Cloned is an iterator that clones the elements of an underlying iterator.
Copied: Copied is an iterator that copies the elements of an underlying iterator.
Empty: Iterator adaptor for the empty() function.
Enumerate: Enumerate is an iterator that returns the current count along with the element. This struct is created by the enumerate() method on IndexedParallelIterator
ExponentialBlocks: ExponentialBlocks is a parallel iterator that consumes itself as a sequence of parallel blocks of increasing sizes (exponentially).
Filter: Filter takes a predicate filter_op and filters out elements that match. This struct is created by the filter() method on ParallelIterator
FilterMap: FilterMap creates an iterator that uses filter_op to both filter and map elements. This struct is created by the filter_map() method on ParallelIterator.
FlatMap: FlatMap maps each element to a parallel iterator, then flattens these iterators together. This struct is created by the flat_map() method on ParallelIterator
FlatMapIter: FlatMapIter maps each element to a serial iterator, then flattens these iterators together. This struct is created by the flat_map_iter() method on ParallelIterator
Flatten: Flatten turns each element to a parallel iterator, then flattens these iterators together. This struct is created by the flatten() method on ParallelIterator.
FlattenIter: FlattenIter turns each element to a serial iterator, then flattens these iterators together. This struct is created by the flatten_iter() method on ParallelIterator.
Fold: Fold is an iterator that applies a function over an iterator producing a single value. This struct is created by the fold() method on ParallelIterator
FoldChunks: FoldChunks is an iterator that groups elements of an underlying iterator and applies a function over them, producing a single value for each group.
FoldChunksWith: FoldChunksWith is an iterator that groups elements of an underlying iterator and applies a function over them, producing a single value for each group.
FoldWith: FoldWith is an iterator that applies a function over an iterator producing a single value. This struct is created by the fold_with() method on ParallelIterator
Inspect: Inspect is an iterator that calls a function with a reference to each element before yielding it.
Interleave: Interleave is an iterator that interleaves elements of iterators i and j in one continuous iterator. This struct is created by the interleave() method on IndexedParallelIterator
InterleaveShortest: InterleaveShortest is an iterator that works similarly to Interleave, but this version stops returning elements once one of the iterators run out.
Intersperse: Intersperse is an iterator that inserts a particular item between each item of the adapted iterator. This struct is created by the intersperse() method on ParallelIterator
IterBridge: IterBridge is a parallel iterator that wraps a sequential iterator.
Map: Map is an iterator that transforms the elements of an underlying iterator.
MapInit: MapInit is an iterator that transforms the elements of an underlying iterator.
MapWith: MapWith is an iterator that transforms the elements of an underlying iterator.
MaxLen: MaxLen is an iterator that imposes a maximum length on iterator splits. This struct is created by the with_max_len() method on IndexedParallelIterator
MinLen: MinLen is an iterator that imposes a minimum length on iterator splits. This struct is created by the with_min_len() method on IndexedParallelIterator
MultiZip: MultiZip is an iterator that zips up a tuple of parallel iterators to produce tuples of their items.
Once: Iterator adaptor for the once() function.
PanicFuse: PanicFuse is an adaptor that wraps an iterator with a fuse in case of panics, to halt all threads as soon as possible.
Positions: Positions takes a predicate predicate and filters out elements that match, yielding their indices.
Repeat: Iterator adaptor for the repeat() function.
RepeatN: Iterator adaptor for the repeat_n() function.
Rev: Rev is an iterator that produces elements in reverse order. This struct is created by the rev() method on IndexedParallelIterator
Skip: Skip is an iterator that skips over the first n elements. This struct is created by the skip() method on IndexedParallelIterator
SkipAny: SkipAny is an iterator that skips over n elements from anywhere in I. This struct is created by the skip_any() method on ParallelIterator
SkipAnyWhile: SkipAnyWhile is an iterator that skips over elements from anywhere in I until the callback returns false. This struct is created by the skip_any_while() method on ParallelIterator
Split: Split is a parallel iterator using arbitrary data and a splitting function. This struct is created by the split() function.
StepBy: StepBy is an iterator that skips n elements between each yield, where n is the given step. This struct is created by the step_by() method on IndexedParallelIterator
Take: Take is an iterator that iterates over the first n elements. This struct is created by the take() method on IndexedParallelIterator
TakeAny: TakeAny is an iterator that iterates over n elements from anywhere in I. This struct is created by the take_any() method on ParallelIterator
TakeAnyWhile: TakeAnyWhile is an iterator that iterates over elements from anywhere in I until the callback returns false. This struct is created by the take_any_while() method on ParallelIterator
TryFold: TryFold is an iterator that applies a function over an iterator producing a single value. This struct is created by the try_fold() method on ParallelIterator
TryFoldWith: TryFoldWith is an iterator that applies a function over an iterator producing a single value. This struct is created by the try_fold_with() method on ParallelIterator
UniformBlocks: UniformBlocks is a parallel iterator that consumes itself as a sequence of parallel blocks of constant sizes.
Update: Update is an iterator that mutates the elements of an underlying iterator before they are yielded.
WalkTree: ParallelIterator for arbitrary tree-shaped patterns. Returned by the walk_tree() function.
WalkTreePostfix: ParallelIterator for arbitrary tree-shaped patterns. Returned by the walk_tree_postfix() function.
WalkTreePrefix: ParallelIterator for arbitrary tree-shaped patterns. Returned by the walk_tree_prefix() function.
WhileSome: WhileSome is an iterator that yields the Some elements of an iterator, halting as soon as any None is produced.
Zip: Zip is an iterator that zips up a and b into a single iterator of pairs. This struct is created by the zip() method on IndexedParallelIterator
ZipEq: An IndexedParallelIterator that iterates over two parallel iterators of equal length simultaneously.

Enums§

Either: The enum Either with variants Left and Right is a general purpose sum type with two cases.

Traits§

FromParallelIterator: FromParallelIterator implements the creation of a collection from a ParallelIterator. By implementing FromParallelIterator for a given type, you define how it will be created from an iterator.
IndexedParallelIterator: An iterator that supports “random access” to its data, meaning that you can split it at arbitrary indices and draw data from those points.
IntoParallelIterator: IntoParallelIterator implements the conversion to a ParallelIterator.
IntoParallelRefIterator: IntoParallelRefIterator implements the conversion to a ParallelIterator, providing shared references to the data.
IntoParallelRefMutIterator: IntoParallelRefMutIterator implements the conversion to a ParallelIterator, providing mutable references to the data.
ParallelBridge: Conversion trait to convert an Iterator to a ParallelIterator.
ParallelDrainFull: ParallelDrainFull creates a parallel iterator that moves all items from a collection while retaining the original capacity.
ParallelDrainRange: ParallelDrainRange creates a parallel iterator that moves a range of items from a collection while retaining the original capacity.
ParallelExtend: ParallelExtend extends an existing collection with items from a ParallelIterator.
ParallelIterator: Parallel version of the standard iterator trait.

Functions§

empty: Creates a parallel iterator that produces nothing.
once: Creates a parallel iterator that produces an element exactly once.
repeat: Creates a parallel iterator that endlessly repeats element (by cloning it). Note that this iterator has “infinite” length, so typically you would want to use zip or take or some other means to shorten it, or consider using the repeat_n() function instead.
repeat_n: Creates a parallel iterator that produces n repeats of element (by cloning it).
repeatnDeprecated: Creates a parallel iterator that produces n repeats of element (by cloning it).
split: The split function takes arbitrary data and a closure that knows how to split it, and turns this into a ParallelIterator.
walk_tree: Create a tree like parallel iterator from an initial root node. The children_of function should take a node and iterate on all of its child nodes. The best parallelization is obtained when the tree is balanced but we should also be able to handle harder cases.
walk_tree_postfix: Create a tree like postfix parallel iterator from an initial root node. The children_of function should take a node and iterate on all of its child nodes. The best parallelization is obtained when the tree is balanced but we should also be able to handle harder cases.
walk_tree_prefix: Create a tree-like prefix parallel iterator from an initial root node. The children_of function should take a node and return an iterator over its child nodes. The best parallelization is obtained when the tree is balanced but we should also be able to handle harder cases.