Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Haskell default io buffering

Yesterday I wrote a little xinetd exercise for my students: make a reverse echo program.

To learn something new, I tried to implement a Haskell solution. The trivial main = forever $ interact reverse does not work. I went through this question and made a corrected version:

import Control.Monad
import System.IO
 
main = forever $ interact revLines
 
revLines = unlines . map (reverse) . lines 

But this corrected version also doesn't work. I read the buffering documentation and played with the various settings. If I set NoBuffering or LineBuffering, my program works correctly. Finally I printed out the default buffering modes for stdin and stdout.

import System.IO

main = do 
  hGetBuffering stdin >>= print 
  hGetBuffering stdout >>= print

I've got BlockBuffering Nothing if I run my program from xinetd(echo "test" | nc localhost 7) but from cli I've got LineBuffering

  • What is the difference between a xinetd tcp service and a cli program, regards to buffering?
  • Do I have to set manually the buffering if I want to write a working program with both running method?

Edit: Thank you all for the helpful answers.

I accept the answer which was given by blaze, he give me a hint with isatty(3). I went through again the System.IO documentation and found the hIsTerminalDevice function, with which I am able to check the connection of the handle.

For the record, here is my final program:

{-# OPTIONS_GHC -W #-}

import System.IO
 
main = do
  hSetBuffering stdin LineBuffering
  hSetBuffering stdout LineBuffering
  
  interact revLines
 
revLines = unlines . map (reverse) . lines 
like image 641
pasja Avatar asked Oct 22 '13 14:10

pasja


2 Answers

It's not specific to Haskell (e.g. the standard C library does the same thing). Traditionally, if a file descriptor corresponds to the terminal, buffering is set to line mode, otherwise to block mode. File descriptor type can be checked by the isatty(3) function -- not sure if it is exported to System.IO.

And yes, you need to set buffering mode manually if you depend on it.

By the way, you can cheat the system and force block buffering in the command line by running your program as cat | ./prog | cat.

like image 139
blaze Avatar answered Oct 21 '22 16:10

blaze


The GHC runtime system tries to be clever when it chooses the default buffering. If it looks like stdin and stdout are directly connected to terminal, they will be line-buffered. If it looks like they are connected to something else, they are block-buffered. This can be problematic if you want to run a program with line-by-line input that doesn't come directly from a terminal. For example, I think that cat | your-program behaves differently from just your-program.

Do i have to set manually the buffering if i want to write a working program with both running method?

Yes.

like image 35
Toxaris Avatar answered Oct 21 '22 18:10

Toxaris