Roadmap

pyfect is in early development. The core effect system and Option type are stable and ready to use. This page outlines what is coming next, roughly in priority order.

Context & Dependency Injection

The R type parameter in Effect[A, E, R] is already part of the type signature but currently always None. The plan is to give it meaning: R declares the services an effect requires to run.

This enables dependency injection without globals — effects declare their requirements in the type, and the runtime ensures they are satisfied before the effect runs. The design will follow Effect's Layer model for composing and providing services.

# Planned API
from pyfect import effect

class Database: ...
class Logger: ...

def find_user(user_id: int) -> effect.Effect[str, NotFoundError, Database]:
    ...

Error Recovery

Practical error handling combinators that are missing from the current release:

• catch — recover from a specific error by returning a new effect
• catch_all — recover from any failure
• retry — re-run an effect with a configurable policy (fixed delay, exponential backoff, max attempts)
• timeout — fail an effect if it does not complete within a given duration

Structured Concurrency & Primitives

The largest milestone. Once fibers exist, the full set of concurrency primitives follows:

• Fiber — a lightweight unit of concurrent execution with structured lifetime
• Cancellation — automatic propagation of cancellation through fiber hierarchies
• Deferred — a single-value async variable, set once and awaited by many
• Queue — a bounded or unbounded async queue for producer/consumer workflows
• PubSub — a broadcast channel for fan-out messaging
• Semaphore — rate-limiting and mutual exclusion
• Latch — a countdown latch for synchronising a group of fibers
• Scope & bracket — structured resource management with guaranteed cleanup on success, failure, or cancellation

Effect Combinators

Quality-of-life combinators for working with collections of effects:

• zip_with — combine two effects into one, failing if either fails
• all — run a list or dict of effects and collect results
• race — run multiple effects and return the first to complete