Closing stdin by pressing Ctrl+D, is a thing you can do during `readline`

[oxinabox]

This has been bothing me for a little while,
This is basically a regression of #1306

Consider the following example code
(It need to be in a block. running each line on a seperate repl prompt will not do this)

begin
       x = readline()
       y = readline()
       @show (x,y)
end

If you press Control+D to sent a EOF to the first readline() then all subsequent calls to readline()will return an empty string. So the output will be(x, y) = ("", "")`

Presumably what is happening is the stdin stream is beng closed by the EOS character.
And reading a line from a closed stream returns "".

This is particularly annoying if you are doing a loop waiting for user input.
E.g. somethilg like

function ask()
   while (true)
    println("Y/N?")
    ans = readline()
    ans == "Y" && return true
    ans == "N" && return false
  end
end

I believe this is why DataDeps tests when a package had not setup the right enviroment variables
would flood the CI with Gigabytes of prompts.
Because CI systems have a closed stdin
( @andreasnoack )

I see a few resolutions, both require special casing the stdin stream

1. Calling readline on stdin, if the stdin stream is closed could throw an error
2. Pressing Ctrl+D in response to to readline, could not cause stdin to be closed

The behavour of Pressing Ctrl+D causing StdIn to close is not documented.
The current behavour of readline on a closed stream is not documented.
which does open the option of documenting this behavour and explaining a work-around.

begin
       x = readline()
       isopen(stdin) || error()
       y = readline()
       isopen(stdin)||errror()
       @show (x,y)
 end

[JeffBezanson]

This is interesting, because readline() does continue to work in subsequent REPL inputs, just not multiple times within the same input.

[oxinabox]

Yes, I assumed this was because the REPL wasn't really using stdin directly.
But was in some sense recreating a new stdin, for each subsequent REPL input.

[vtjnash]

Yes, typing the end-of-transmission character ^D causes us to set the end-of-file bit on the stream, but doesn't actually completely close it. But the user can then choose to reset that bit if desired (as the REPL does).

[JeffBezanson]

The docstring for readline doesn't mention EOF behavior, so we should add that. You need to either pass keep=true and look for the empty string, or check eof(io). So I think this is mostly a doc issue --- the caller needs to check for EOF. Throwing an error is an option, but would be very breaking of course. We can also consider exporting reseteof.