Parallel

Crystal library for parallel processing using Fiber::ExecutionContext.

Requirements

• Crystal 1.19.1+
• Compile flags: -Dpreview_mt -Dexecution_context

Installation

Add to shard.yml:

dependencies:
  parallel:
    github: kojix2/parallel

Run shards install

Usage

require "parallel"

# Parallel map
[1, 2, 3, 4].par_map { |x| x * 2 }
# => [2, 4, 6, 8]

# Parallel each
[1, 2, 3, 4].par_each { |x| puts x }

# Chunk processing (fewer context switches)
[1, 2, 3, 4].par_map(chunk: 2) { |x| x * 2 }
# => [2, 4, 6, 8] (same result, fewer context switches)

# Custom ExecutionContext
context = Fiber::ExecutionContext::Parallel.new("workers", 8)
[1, 2, 3, 4].par_map(context) { |x| x * 2 }

By default, a dedicated Fiber::ExecutionContext::Parallel is created lazily. You may supply your own context if preferred.

# Use system default ExecutionContext
Parallel.execution_context = Fiber::ExecutionContext.default
[1, 2, 3, 4].par_map { |x| x * 2 }

Compilation

crystal build -Dpreview_mt -Dexecution_context your_app.cr
crystal spec -Dpreview_mt -Dexecution_context

Methods

Parallel (global)

• execution_context : Fiber::ExecutionContext::Parallel

  • Returns the library default context.

• execution_context=(context : Fiber::ExecutionContext::Parallel)

  • Sets the library default context.

Enumerable

• par_map(execution_context = nil, *, chunk = nil, &block)

  • Applies block to each element in parallel, returns array of results.
  • chunk: Process elements in chunks to reduce context switches.

• par_each(execution_context = nil, *, chunk = nil, &block)

  • Applies block to each element in parallel for side effects.
  • chunk: Process elements in chunks to reduce context switches.

Notes

• Works with any Enumerable (Array, Range, Set, Hash, etc.)
• Indexable types (Array, Slice) preserve order
• Exceptions are propagated from parallel tasks
• Uses a shared Fiber::ExecutionContext::Parallel by default for performance (configurable via Parallel.execution_context)
• Default worker count is based on Fiber::ExecutionContext.default_workers_count (affected by CRYSTAL_WORKERS or CPU count)
• Adaptive chunking clamps the computed chunk size between 1 and 1000
• Thread safety is your responsibility when accessing shared resources

Contributing

• Fork this repository
• Report bugs
• Fix bugs and submit pull requests
• Write, clarify, or fix documentation
• Suggest or add new features

License

MIT

This library contains code generated by AI.