I have a zip file, containing CSVs, on a remote S3 cellar.

Using Unzip, I can get a stream of the CSV file that I would like to decode, like so :

    aws_s3_config =
      ExAws.Config.new(:s3,
        access_key_id: ["xxx", :instance_role],
        secret_access_key: ["xxx", :instance_role]
      )

    file = new(zip_name, bucket_name, aws_s3_config)
    {:ok, unzip} = Unzip.new(file)
    stream = Unzip.file_stream!(unzip, file_name)

as explained in the doc.

Now I would like to consume that stream by reading it with CSV.
So I try stream |> CSV.decode |> Enum.take(1) and get an error ** (FunctionClauseError) no function clause matching in CSV.Decoding.Preprocessing.Lines.starts_sequence?/5

If I write the content of my CSV on the disk and then read it, it works fine :

# write the file on disk
stream |> Stream.into(File.stream!("stops.txt")) |> Stream.run()
# then read and decode it
File.stream!("stops.txt") |> CSV.decode |> Enum.take(1)

I get the desired result, the first row of the CSV file : [ok: ["\uFEFFstop_id", "stop_name", "stop_lat", "stop_lon", "location_type"]]

The difference I see is that Unzip.file_stream! and File.stream!("stops.txt") do not stream the file the same way. Unzip seem to do it by chunks of 65k, while File.stream! streams line by line.

How can I solve this, without writing the file to disk as an intermediary step?
Thanks!