Close() on zstd encoder doesn't call close on underlying writer.

[cloudshiftchris]

When calling .Close() on a zstd Encoder (via zstd.NewWriter, wrapping an existing writer), it does not call .Close() on the wrapped writer. This requires tracking the underlying writer to ensure it gets closed properly.

Especially problematic when using with io.Pipe (or other streaming situations), as the underlying PipeWriter must be closed to signal to the reader that everything is completed.

The documentation for zstd.NewWriter is silent on this; it provides no indication of who is responsible for lifecycle management of the passed-in Writer.

[LeGEC]

just to mention: this behavior is actually aligned with the compressors in the standard library: compress/gzip, compress/zlib etc ...

The sentence "It does not close the underlying io.Writer." should probably be added in the documentation.

[LeGEC]

Regarding the requirement "I must call .Close() on the underlying object":

Most of the time, I can actually add defer out.Close() (where out is the underlying object) at a relevant spot in my code.

In situations where I needed to explicitly chain several calls to Close, I found myself doing variations on something like:

type writeCloser struct{
    io.Writer
    fnClose func() error
}

func (wc *writeCloser) Close() error {
  if wc.fnClose == nil {
    return nil
  }
  return wc.fnClose()
}

// sample usage:
fnClose = func() error {
   err1 := w1.Close()
   err2 := w2.Close()
   return errors.Join(err1, err2)
}

wc := &writeCloser{w1, fnClose}

Illustrated example: https://go.dev/play/p/SIbVEu6eh7i