fluentfp

README

fluentfp

Fluent functional programming for Go.

Type-safe collection chains, composable resilience (retry, circuit breaker, throttle), bounded concurrency pipelines, typed HTTP handlers, and optional/result types — all on standard Go, no framework required.

See pkg.go.dev for API docs and the showcase for 16 before/after rewrites from real GitHub projects.

Zero reflection. Zero global state. Zero build tags.

Quick Start

Requires Go 1.26+.

go get github.com/binaryphile/fluentfp

import "github.com/binaryphile/fluentfp/slice"

// Before: scaffolding around one line of intent
var names []string                         // state
for _, u := range users {                  // iteration
    if u.IsActive() {                      // predicate
        names = append(names, u.Name())    // accumulation
    }
}

// After: intent only
names := slice.From(users).KeepIf(User.IsActive).ToString(User.Name)

That's a fluent chain — each step returns a value you can call the next method on, so the whole pipeline reads as a single expression: filter, then transform.

Method Expressions

Go lets you reference a method by its type name, creating a function value where the receiver becomes the first argument:

func (u User) IsActive() bool  // method
func(User) bool                // method expression: User.IsActive

KeepIf expects func(T) bool — User.IsActive is exactly that:

names := slice.From(users).KeepIf(User.IsActive).ToString(User.Name)

Without method expressions, every predicate needs a wrapper: func(u User) bool { return u.IsActive() }.

For []*User slices, the method expression is (*User).IsActive.

See naming patterns for when to use method expressions vs named functions vs closures.

What It Looks Like

The showcase has 16 more, including Sort, Trim, and Map-to-Slice.

Conditional Struct Fields

Go struct literals let you build and return a value in one statement — fluentfp keeps it that way when fields are conditional.

Before	After
var level string if overdue { level = "critical" } else { level = "info" } var icon string if overdue { icon = "!" } else { icon = "✓" } return Alert{ Message: msg, Level: level, Icon: icon, }	return Alert{ Message: msg, Level: option.When(overdue, "critical").Or("info"), Icon: option.When(overdue, "!").Or("✓"), }

Go has no inline conditional expression. option.When fills that gap — each field resolves in place, so the struct literal stays a single statement. From hashicorp/consul.

Bounded Concurrent Requests

Fetch weather for a list of cities with at most 10 simultaneous goroutines.

Before — errgroup (21 lines)	After (1 line)
func Cities(ctx context.Context, cities ...string) ([]*Info, error) { g, ctx := errgroup.WithContext(ctx) g.SetLimit(10) res := make([]*Info, len(cities)) for i, city := range cities { g.Go(func() error { info, err := City(ctx, city) if err != nil { return err } res[i] = info return nil }) } if err := g.Wait(); err != nil { return nil, err } return res, nil }	func Cities(ctx context.Context, cities ...string) ([]*Info, error) { return slice.FanOutAll(ctx, 10, cities, City) }

FanOutAll is all-or-nothing: on first error it cancels remaining work and returns that error. City passes directly — no wrapper needed. Panics in callbacks are recovered as *rslt.PanicError with stack trace.

When you need per-item outcomes instead of all-or-nothing, use FanOut:

results := slice.FanOut(ctx, 10, cities, City)
infos, errs := rslt.CollectOkAndErr(results)  // gather successes and failures separately

From the errgroup pattern.

Why fluentfp

Type-safe end-to-end. go-linq gives you []any back — cast it and hope you got the type right. lo requires func(T, int) callbacks, so every stdlib function needs a wrapper to discard the unused index. fluentfp uses generics throughout: Mapper[T] is []T with methods. If it compiles, you avoid a class of cast, index, and callback-shape mistakes.

Fewer places for bugs to hide. No index means no off-by-one in a predicate. No loop variable means no shadowing. No accumulator means no forgetting to initialize one. These are the loop-scaffolding bug classes that code review catches regularly — fluentfp removes the scaffolding where they live.

Works with Go, not against it. Mappers are slices — callers of your functions don't need to import fluentfp. Options use comma-ok (.Get() (T, bool)), the same pattern as map lookups and type assertions. either.Fold gives you exhaustive dispatch the compiler enforces — miss a branch and it doesn't compile. must.BeNil makes invariant enforcement explicit. Mutation, channels, and hot paths stay as loops.

Interchangeable Types

Mapper[T] is defined as type Mapper[T any] []T — a defined type, not a wrapper. []T and Mapper[T] convert implicitly in either direction, so you choose how much to expose:

// Public API — hide the dependency. Callers never see fluentfp types.
func ActiveNames(users []User) []string {
    return slice.From(users).KeepIf(User.IsActive).ToString(User.Name)
}

// Internal — embrace it. Accepting Mapper saves From() calls across a chain of helpers.
func transform(users slice.Mapper[User]) slice.Mapper[User] {
    return users.KeepIf(User.IsActive).Convert(User.Normalize)
}

The public pattern keeps fluentfp as an implementation detail — callers don't import it, and its types don't appear in intellisense. Internal code can pass Mappers between helpers to avoid repeated From() wrapping.

Performance

A single chain step (filter OR map) matches a hand-written loop — slice.From is a zero-cost type conversion, and each operation pre-allocates with make([]T, 0, len(input)).

Multi-step chains pay one allocation per stage. A chain that filters then maps makes two passes and two allocations where a hand-written loop can fuse them into one. In benchmarks (1000 elements), a two-step chain runs ~2.5× slower than the fused equivalent.

If you're counting nanoseconds in a hot path, fuse it in a loop. Most loops aren't hot paths — they're scaffolding that fluentfp eliminates.

Measurable Impact

Codebase Type	Code Reduction	Complexity Reduction
Mixed (typical)	12%	26%
Pure pipeline	47%	95%

Individual loops see up to 6x line reduction. Codebase-wide averages are lower because not every line is a loop. Complexity measured via scc. See methodology.

When to Use Loops

fluentfp replaces mechanical loops — iteration scaffolding around a predicate and a transform. It doesn't try to replace loops that do structural work:

// Mutation in place — fluentfp returns new slices, but elements are shared (shallow copy)
for i := range items {
    if items[i].ID == target {
        items[i].Status = "done"
        break
    }
}

// Channel consumption — direct range is simplest for straightforward use
for msg := range ch {
    handle(msg)
}

// Bridge to fluentfp when you want operators on a channel
seq.FromChannel(ctx, ch).KeepIf(valid).Take(10).Each(handle)

// Complex control flow — early return, labeled break
for _, item := range items {
    if item.IsTerminal() {
        return item.Result()
    }
}

// Hot paths — fuse filter+map into one pass when nanoseconds matter
out := make([]R, 0, len(input))
for _, v := range input {
    if keep(v) {
        out = append(out, transform(v))
    }
}

Packages

Packages are independent — import one or all.

Package	Purpose	Key Functions
slice	Collection transforms	KeepIf, RemoveIf, Fold, FanOutAll
kv	Map transforms	KeepIf, MapValues, Map, Values
seq	Fluent iter.Seq chains	From, KeepIf, Take, Collect
stream	Lazy memoized sequences	Generate, Unfold, Take, Collect
option	Optional values + conditionals	Of, When, Or, NonZero, Env
either	Sum types	Left, Right, Fold, Convert, FlatMap
rslt	Typed error handling	Ok, Err, CollectAll, CollectOkAndErr
must	Invariant enforcement	Get, BeNil, Of
hof	Resilience + function combinators	Retry, WithBreaker, Throttle, MapErr
toc	Bounded pipeline stages	Start, Pipe, NewBatcher, NewTee
ctxval	Typed context values	With, From, NewKey
web	Typed HTTP handlers	Adapt, DecodeJSON, Steps
memo	Memoization	Of, Fn, FnErr, NewLRU
heap	Persistent priority queue	New, Insert, Pop, Collect
pair	Zip slices	Zip, ZipWith
combo	Combinatorial constructions	CartesianProduct, Combinations, PowerSet
lof	Lower-order function wrappers	Len, Println, Identity, Inc

Package Highlights

hof — composable resilience decorators:

// Retry with exponential backoff, only for transient errors
backoff := hof.ExponentialBackoff(100 * time.Millisecond)
fetcher := hof.Retry(3, backoff, isTransient, fetchData)

// Circuit breaker — trips after 5 consecutive failures, resets after 30s
cfg := hof.BreakerConfig{ResetTimeout: 30 * time.Second}
breaker := hof.NewBreaker(cfg)
protected := hof.WithBreaker(breaker, fetcher)

// All decorators share func(ctx, T) (R, error) — stack freely
throttled := hof.Throttle(10, protected)

web — typed HTTP handlers on net/http:

// Handlers return Result[Response] — no ResponseWriter, no manual status codes
var createUser web.Handler = func(r *http.Request) rslt.Result[web.Response] {
    req := rslt.Of(web.DecodeJSON[CreateReq](r))
    return rslt.Map(req, createAndRespond)
}

// Adapt bridges to http.HandlerFunc; WithErrorMapper translates domain errors
endpoint := web.Adapt(createUser, web.WithErrorMapper(domainToHTTP))
mux.HandleFunc("POST /users", endpoint)

ctxval — typed context values without type assertions:

type RequestID string
ctx = ctxval.With(ctx, RequestID("abc-123"))
reqID := ctxval.From[RequestID](ctx)  // Option[RequestID]

rslt — typed error handling as values:

r := rslt.Of(strconv.Atoi(input))  // wrap (int, error) → Result[int]
port := r.Or(8080)                   // value or default

seq — fluent chains on Go's iter.Seq:

active := seq.FromIter(maps.Keys(configs)).KeepIf(isActive).Collect()

stream — lazy memoized sequences:

naturals := stream.Generate(0, lof.Inc)
first10Squares := stream.Map(naturals, square).Take(10).Collect()

Capability Map

If you need to...	Use	Package
Filter, map, or fold a slice	slice.From(s).KeepIf(f).ToString(g)	slice
Run work concurrently with a limit	slice.FanOutAll(ctx, 10, items, fn)	slice
Retry on failure with backoff	hof.Retry(3, backoff, shouldRetry, fn)	hof
Circuit-break an unhealthy dependency	hof.WithBreaker(breaker, fn)	hof
Throttle concurrent access	hof.Throttle(n, fn)	hof
Transform errors in a decorator chain	hof.MapErr(fn, mapper)	hof
Debounce rapid calls	hof.NewDebouncer(wait, fn)	hof
Represent optional values	option.Of(v), option.NonZero(v), option.Env("KEY")	option
Inline conditional (no ternary in Go)	option.When(cond, val).Or(fallback)	option
Handle (value, error) as a single value	rslt.Of(strconv.Atoi(s))	rslt
Collect per-item outcomes from FanOut	rslt.CollectOkAndErr(results)	rslt
Exhaustive two-branch dispatch	either.Fold(e, onLeft, onRight)	either
Panic on invariant violation	must.Get(fn()), must.BeNil(err)	must
Store typed values in context.Context	ctxval.With(ctx, val) / ctxval.From[T](ctx)	ctxval
Build typed HTTP handlers on net/http	web.Adapt(handler, web.WithErrorMapper(m))	web
Decode JSON request bodies	web.DecodeJSON[T](r)	web
Run a bounded pipeline with backpressure	toc.Start → toc.Pipe → toc.Pipe	toc
Batch items by count or weight	toc.NewBatcher(ctx, src, n)	toc
Broadcast to N branches	toc.NewTee(ctx, src, n)	toc
Lazy iterate with memoization	stream.Generate(seed, fn).Take(10).Collect()	stream
Lazy iterate without memoization	seq.From(s).KeepIf(f).Take(10).Collect()	seq
Memoize a function	memo.Of(fn) or memo.Fn(cache, fn)	memo
Work with maps functionally	kv.Keys(m), kv.MapValues(m, fn)	kv
Generate combinations/permutations	combo.Combinations(items, k)	combo
Use a persistent priority queue	heap.New(cmp).Insert(v)	heap
Zip two slices into pairs	pair.Zip(as, bs) or pair.ZipWith(as, bs, fn)	pair

Further Reading

• Full Analysis - Technical deep-dive with examples
• Methodology - How claims were measured
• Nil Safety - The billion-dollar mistake and Go
• Naming Functions - Function naming patterns for HOF use
• Library Comparison - How fluentfp compares to alternatives
• Real-World Showcase - Before/after rewrites from GitHub projects

See CHANGELOG for version history.

License

fluentfp is licensed under the MIT License. See LICENSE for more details.