A safe way in Ruby to download a file to disk using open-uri (with/without comments)

01-safe-download.rb

require "open-uri"
require "net/http"

Error = Class.new(StandardError)

DOWNLOAD_ERRORS = [
  SocketError,
  OpenURI::HTTPError,
  RuntimeError,
  URI::InvalidURIError,
  Error,
]

def download(url, max_size: nil)
  url = URI.encode(URI.decode(url))
  url = URI(url)
  raise Error, "url was invalid" if !url.respond_to?(:open)

  options = {}
  options["User-Agent"] = "MyApp/1.2.3"
  options[:content_length_proc] = ->(size) {
    if max_size && size && size > max_size
      raise Error, "file is too big (max is #{max_size})"
    end
  }

  downloaded_file = url.open(options)

  if downloaded_file.is_a?(StringIO)
    tempfile = Tempfile.new("open-uri", binmode: true)
    IO.copy_stream(downloaded_file, tempfile.path)
    downloaded_file = tempfile
    OpenURI::Meta.init downloaded_file, stringio
  end

  downloaded_file

rescue *DOWNLOAD_ERRORS => error
  raise if error.instance_of?(RuntimeError) && error.message !~ /redirection/
  raise Error, "download failed (#{url}): #{error.message}"
end

02-safe-download-commented.rb

require "open-uri"
require "net/http" # Just to get the SocketError class

Error = Class.new(StandardError)

DOWNLOAD_ERRORS = [
  SocketError,          # domain not found
  OpenURI::HTTPError,   # response status 4xx or 5xx
  RuntimeError,         # redirection errors (e.g. redirection loop)
  URI::InvalidURIError, # invalid URL
  Error,                # our errors
]

def download(url, max_size: nil)
  # URLs with spaces will raise an InvalidURIError, so we need to encode it.
  # However, the user can pass an already encoded URL, so we first need to
  # decode it.
  url = URI.encode(URI.decode(url))

  # This will raise an InvalidURIError if the URL is very wrong. It will still
  # pass for strings like "foo", though.
  url = URI(url)

  # We need to check if the URL was either http://, https:// or ftp://, because
  # these are the only ones we can download from. open-uri will add the #open
  # method only to these ones, so this is a good check.
  raise Error, "url was invalid" if !url.respond_to?(:open)

  options = {}
  # It was shown that in a random sample approximately 20% of websites will
  # simply refuse a request which doesn't have a valid User-Agent.
  options["User-Agent"] = "MyApp/1.2.3"
  # It's good to shield ourselves from files that are too big. open-uri will
  # call this block as soon as it gets the "Content-Length" header, which means
  # that we can bail out before we download the file.
  options[:content_length_proc] = ->(size) {
    if max_size && size && size > max_size # sometimes "Content-Length" can be empty
      raise Error, "file is too big (max is #{max_size})"
    end
  }

  # Finally we download the file. Here we mustn't use simple #open that open-uri
  # overrides, because this is vulnerable to shell execution attack (if #open
  # method detects a starting pipe (e.g. "| ls"), it will execute the following
  # as a shell command).
  downloaded_file = url.open(options)

  # open-uri will return a StringIO instead of a Tempfile if the filesize
  # is less than 10 KB, so we patch this behaviour by converting it into a
  # Tempfile.
  if downloaded_file.is_a?(StringIO)
    # We need to open it in binary mode for Windows users.
    tempfile = Tempfile.new("open-uri", binmode: true)
    # IO.copy_stream is the most efficient way of data transfer.
    IO.copy_stream(downloaded_file, tempfile.path)
    downloaded_file = tempfile
    # We add the metadata that open-uri puts on the file (e.g. #content_type)
    OpenURI::Meta.init downloaded_file, stringio
  end

  downloaded_file # Finally

rescue *DOWNLOAD_ERRORS => error
  # open-uri will throw a RuntimeError when it detects a redirection loop, so
  # we want to reraise the exception if it was some other RuntimeError
  raise if error.instance_of?(RuntimeError) && error.message !~ /redirection/
  # We raise our unified Error class
  raise Error, "download failed (#{url}): #{error.message}"
end

I can see that url.open downloads the whole file to memory first.

This is crap