well-typed / borsh Goto Github PK

decodeLitleEndian should not have been exported, and decodeLargeToken, decodeIncremental and decodeIncremental_ should get some docs. When we do, is there still a need to expose the internals of Decoder..? What is the minimal thing we should export? 🤔

It should not be possible to coerce say FixedSizeArray 32 Word8 to FixedSizeArray 64 Word8.

On the rust side, a struct (keyword struct) will not have a tag, but the general case (keyword enum) will get a tag. On the Haskell side we not have different keywords for this, and a struct is simply an ADT with a single constructor. However, this means that if we derive FromBorsh and co for such a struct, the decoder will expect a tag, but the Rust-side encoder won't provide one. We can recover the right behaviour by using deriving .. via Struct MyStruct, but this is easy to forget. Perhaps we shoud force users to always use deriving via, deriving .. via Struct T versus deriving .. via Enum T.

We don't need the full footprint of memory.

See

• https://hackage.haskell.org/package/base-4.12.0.0/docs/GHC-ByteOrder.html
• https://hackage.haskell.org/package/base-4.12.0.0/docs/GHC-Exts.html#v:byteSwap-35-.
• https://hackage.haskell.org/package/base-4.12.0.0/docs/Data-Word.html#v:byteSwap64
• https://hackage.haskell.org/package/wide-word-0.1.1.2/docs/Data-WideWord-Word128.html#v:byteSwapWord128