wireform-flatbuffers

wireform-flatbuffers implements Google FlatBuffers, a zero-copy serialization format designed for game engines, mobile apps, and real-time inference serving. FlatBuffers lay out tables with vtables that index field offsets, keeping scalars inline and strings or sub-tables behind indirection. Use this package when you need to read large buffers without deserializing the entire message, or when sharing the builder infrastructure with Arrow IPC.

Key features

• Typeclass API via ToFlatBuffers and FromFlatBuffers with Template Haskell deriving
• FlatBuffers IDL parser and codegen from .fbs schema files
• Vtable-based wire layout with inline scalars and offset-indirected collections
• Zero-copy view/reader via FlatBuffers.View for schema-known access patterns
• Shared builder with Arrow IPC for cross-format buffer construction
• QuasiQuoter for inline [flatbuffers| ... |] schemas

Basic usage

Derive instances for your table types, then encode and decode through the value-level codec:

{-# LANGUAGE DerivingStrategies #-}
{-# LANGUAGE TemplateHaskell #-}

import FlatBuffers.Decode qualified as FBD
import FlatBuffers.Derive (ToFlatBuffers (..), FromFlatBuffers (..), deriveFlatBuffers, deriveView)
import FlatBuffers.Encode qualified as FBE
import Data.Int (Int32)
import Data.Text (Text)
import GHC.Generics (Generic)

data Widget = Widget
  { widgetName  :: !Text
  , widgetCount :: !Int32
  , widgetPrice :: !Double
  }
  deriving stock (Show, Eq, Generic)

$(deriveFlatBuffers ''Widget)
$(deriveView ''Widget)

encodeWidget :: Widget -> ByteString
encodeWidget = FBE.encode . toFlatBuffers

decodeWidget :: ByteString -> Either String Widget
decodeWidget bs = do
  val <- FBD.decode bs
  fromFlatBuffers val

sample :: Widget
sample = Widget "wireform" 42 2.718

roundTrip :: Either String Widget
roundTrip = decodeWidget (encodeWidget sample)

For zero-copy reads on known schemas, use the view layer after encoding:

import FlatBuffers.View (decodeRoot)

readWidget :: ByteString -> Either String Widget
readWidget = decodeRoot

Generate types from .fbs files:

{-# LANGUAGE TemplateHaskell #-}
import FlatBuffers.QQ (fbs)

[fbs|
  table Widget {
    name:string;
    count:int;
    price:double;
  }
  root_type Widget;
|]

wireform-gen flatbuffers -i schema.fbs -o src/Gen/

Performance

Encode/decode (zero-copy decode)

Operation	encode	decode	ratio
Person table	827 ns	142 ns	0.17x
Person[100] vector	95352 ns	75.7 ns	0.00x

_{Last run 2026-06-27 11:35:54 UTC. ghc-9.8.4 on darwin-aarch64, criterion 1.6.5. Decode is a zero-copy cursor by design: only the outer envelope is resolved at decode time. Per-field reads happen lazily..}

Like Cap’n Proto, FlatBuffers decode is a zero-copy cursor. The decode cost is near-constant because only the root table offset is resolved; field access is lazy pointer arithmetic into the original buffer. Encode is proportional to the number of fields and vector elements.

The chart and table above are regenerated by wireform-stats from wireform-flatbuffers/bench-results/summary/flatbuffers-encode-decode.json — the same source the README chart is built from.

Notable modules

Module	Purpose
FlatBuffers.Derive	ToFlatBuffers / FromFlatBuffers and deriveFlatBuffers
FlatBuffers.Encode / FlatBuffers.Decode	High-level encoder and value-tree decoder
FlatBuffers.View	Zero-copy cursor access for schema-known tables
FlatBuffers.Reader	Low-level pointer-walking decoder
FlatBuffers.Builder	Vtable and offset builder
FlatBuffers.Value	Dynamic untyped Value ADT
FlatBuffers.Schema / FlatBuffers.Parser	Schema AST and .fbs parser
FlatBuffers.CodeGen / FlatBuffers.QQ	Haskell codegen and quasiquoter
FlatBuffers.Registry	Runtime schema registry