add position()

[alyst]

It would be nice to have Base.position(io::TranscodingStream) to return the current position in the transcoded stream.
There's already state.total_out, so it should not be impossible to keep track of the transcoded bytes since the last reset (+myabe adjust it w.r.t unread()).

The use case.
I came across this issue once again when trying to convert MLDatasets.jl to use CodecZlib.jl instead of the aging GZip.jl.
The gzipped files there are collections of images, each image occupying the same size. So, given image index, we can tell its exact position in the decoding stream.
When implementing readimages(io, image_indices::AbstractVector), it was natural to define readimage(io, image_index).
To work, readimage() just needs to know the current stream position, so that it can calculate how many bytes to skip to get to the specified image.
Without position(io) one would have to keep track of this information somewhere externally, which looks like an unnecessary complication to me.

[damiendr]

I would also like having position() on transcoding streams, so that I could use them with a parser that does something like this:

len = read(io, UInt32)
offset = position(io)
while position(io) < offset + len
   read_item(io)
end

[zgornel]

position seems to work for the Noop() stream however one has to flush the stream first:

julia> using TranscodingStreams, Serialization
       fid = TranscodingStream(Noop(), open("./__deleteme__.bin","w"))
       serialize(fid, 1)
       @show position(fid)
       flush(fid)
       @show position(fid)
# position(fid) = -9
# position(fid) = 9

[laborg]

Looking at CSV.jl again, it appears that implementing seek would be necessary too.

@bicycle1885 : I don't want to pressure you, but do you know when you will find time? With this information dependent packages could easier decide whether or not to create an intermediate work around. (e.g. load everything into a IOBuffer for RDatasets).