From: "akr (Akira Tanaka) via ruby-core" <ruby-core@...>
Date: 2024-04-17T23:16:04+00:00
Subject: [ruby-core:117580] [Ruby master Feature#20215] Introduce `IO#readable?`

Issue #20215 has been updated by akr (Akira Tanaka).


Thank you for the explanation of the motivation.
I feel it is reasonable.

However, mame-san still had a question yesterday: why don't you detect an error in writing a request?
I guess it is because writing the request may not fail.
It needs two writes to detect the error.
(The first write in a client application causes the client kernel to send a packet to the server.
The server kernel responds to it with a RST packet because the server socket is closed.
The client kernel receives the RST packet and detects we cannot send data in the connection.
The second write in the client application will cause an error.)
Please point out if I am wrong.

My next question is why `io.eof?` and `io.wait_readable(0)` doesn't fulfill your motivation.
My current guess is related to the buffering mechanism of IO class.
But I'm not certain.





----------------------------------------
Feature #20215: Introduce `IO#readable?`
https://bugs.ruby-lang.org/issues/20215#change-107980

* Author: ioquatix (Samuel Williams)
* Status: Open
----------------------------------------
There are some cases where, as an optimisation, it's useful to know whether more data is potentially available.

We already have `IO#eof?` but the problem with using `IO#eof?` is that it can block indefinitely for sockets.

Therefore, code which uses `IO#eof?` to determine if there is potentially more data, may hang.

```ruby
def make_request(path = "/")
  client = connect_remote_host
  # HTTP/1.0 request:
  client.write("GET #{path} HTTP/1.0\r\n\r\n")

  # Read response
  client.gets("\r\n") # => "HTTP/1.0 200 OK\r\n"

  # Assuming connection close, there are two things the server can do:
  # 1. peer.close
  # 2. peer.write(...); peer.close

  if client.eof? # <--- Can hang here!
    puts "Connection closed"
    # Avoid yielding as we know there definitely won't be any data.
  else
    puts "Connection open, data may be available..."
    # There might be data available, so yield.
    yield(client)
  end
ensure
  client&.close
end

make_request do |client|
  puts client.read # <--- Prefer to wait here.
end
```

The proposed `IO#readable?` is similar to `IO#eof?` but rather than blocking, would simply return false. The expectation is the user will subsequently call `read` which may then wait.

The proposed implementation would look something like this:

```ruby
class IO
  def readable?
    !self.closed?
  end
end

class BasicSocket
  # Is it likely that the socket is still connected?
  # May return false positive, but won't return false negative.
  def readable?
    return false unless super
    
    # If we can wait for the socket to become readable, we know that the socket may still be open.
    result = self.recv_nonblock(1, MSG_PEEK, exception: false)
    
    # No data was available - newer Ruby can return nil instead of empty string:
    return false if result.nil?
    
    # Either there was some data available, or we can wait to see if there is data avaialble.
    return !result.empty? || result == :wait_readable
    
  rescue Errno::ECONNRESET
    # This might be thrown by recv_nonblock.
    return false
  end
end
```

For `IO` itself, when there is buffered data, `readable?` would also return true immediately, similar to `eof?`. This is not shown in the above implementation as I'm not sure if there is any Ruby method which exposes "there is buffered data".



-- 
https://bugs.ruby-lang.org/
 ______________________________________________
 ruby-core mailing list -- ruby-core@ml.ruby-lang.org
 To unsubscribe send an email to ruby-core-leave@ml.ruby-lang.org
 ruby-core info -- https://ml.ruby-lang.org/mailman3/postorius/lists/ruby-core.ml.ruby-lang.org/