Subj : Re: EOF (end of file) problem
To   : borland.public.cpp.borlandcpp
From : maeder@glue.ch (Thomas Maeder [TeamB])
Date : Sat Nov 01 2003 01:12 pm

Karlo Basic <karlman@email.si> writes:

> I hope I'm asking the right question in the right group.

Strictly speaking, you don't. But I refrain from my usual response this time,
because your question is relevant to Borland C++ (to whom this newsgroup is
on topic) users as well to C++ Builder users like you.


> I have a problem with a piece of C++ code that runs well under Linux
> (gcc 2.96) but doesn't run as expected in Windows (Borland C++ 5.5.1):

Borland C++ 5.5.1 is what the C++ Builder 5 command line compiler calls
itself, somewhat misleadingly IMHO.

BTW: There was never a gcc 2.96; the latest 2.xx releases all had 2.95.x.
2.96 is just a RedHat hack.


> Here's the code:
> **********
> int num;
> ifstream in_file;
> in_file.open("whatever.txt");
> while (!in_file.eof())
>      cin >> num;
> **********

This code is not correct.

First of all, I believe you mean the last line to read

in_file >> num;

, or you'll have an infinite loop. Please take care to always copy&paste
minimal code as you have tested it; don't re-type.


Second, the only way the while loop can stop is when the eof flag inside
in_file is set. Now input from in_file can fail for other reasons than end
of file; once this happens, you'll have an infinite loop, as well, because
the flags inside in_file aren't going to change any more.

A correct version of your program looks like this:

#include <fstream>
#include <iostream>

int main()
{
  std::ifstream in_file("whatever.text");
  int num;
  while (in_file >> num)
    std::cout << num << '\n';
}

[I've added the output line to make testing easier]

Note to Borland C++ users: In Borland C++, <fstream.h> and <iostream.h>
might have to be #included, and the std:: qualifications removed.


The big differences between our programs are:

- mine checks for failure only after the read attempt; stream flags such as
  end of file and input failure are only set while reading, never before
  reading

- the loop stops after any kind of input failure has been met: file can't be
  opened, bad input, end of file, you name it


> Example input:
> 1 2 3 4 5 6 7EOF -> Linux OK; Windows not
> 1 2 3 4 5 6 7
> EOF         -> Linux & Windows OK

I think that this difference should go away in my program.

.