Unusual Encoder/Decoder signatures?

[dmitshur]

Please correct me if I'm wrong. I don't think there's a standard interface for Encoder/Decoder types, but as far as I can see, they all typically take an io.Reader to create a Decoder, and io.Writer to create an Encoder.

For example, look at:

http://godoc.org/encoding/json#NewDecoder
http://godoc.org/encoding/xml#NewDecoder
http://godoc.org/encoding/ascii85#NewDecoder

Then, their Encode/Decode methods take one parameter and return just error.

The Encoder/Decoder in this package do not take a writer/reader, but instead return the encoded/decoded data alongside the error.

My question is, would it be a better API if they did take a Reader/Writer?

[dmitshur]

You can see what prompted this question if you look at the contrast in this snippet of code:

encodeUsingMafsa := func(w io.Writer) error {
    data, err := new(mafsa.Encoder).Encode(bt)
    if err != nil {
        return err
    }
    _, err = w.Write(data)
    return err
}

encodeUsingGob := func(w io.Writer) error {
    return gob.NewEncoder(w).Encode(ss)
}

However, I now realize I could've relied on the encoding.BinaryMarshaler interface which *mafsa.BuildTree implements instead of the encoder. But then I would need something that implements encoding.BinaryMarshaler for a slice of strings...

[mholt]

Yes, it definitely would be better. I did make some effort to implement binary Marshaler and Unmarshaler, but didn't add Reader/Writer support (yet?).

I actually do not know a lot about io.Reader/Writer (I know, I know, a heinous ignorance in the Go community) and so I haven't given it much thought. I would certainly like them to be able to use io.Reader/Writer interfaces, but I just need to learn more about how they work. For example, decoding can't be done unless the entire byte stream is available. Can an io.Reader guarantee that? (Each edge represented in binary has a pointer to another part of the file, which could be anywhere.) I guess all this package would just have to do is a ReadAll to get the entire byte stream. You think that might work?

[dmitshur]

For example, decoding can't be done unless the entire byte stream is available. Can an io.Reader guarantee that? (Each edge represented in binary has a pointer to another part of the file, which could be anywhere.) I guess all this package would just have to do is a ReadAll to get the entire byte stream. You think that might work?

In that case, there's not winning much from using Encoder/Decoder, except a more familiar API.

That should be just fine. If you need the entire stream to be available, then get it using ReadAll, which would guarantee you've read until the stream ended (io.EOF) if the error returned from ReadAll is nil.