OJ Board

Note: I have edited this post to remove an incorrect claim about getline(istream, string).

Recently a friend ran into difficulties with a WA on a simple problem. We tried many things (understatement) before finally discovering that the difficulty lay in his input routine:

[cpp]
// one version which didn't work
for (;;) {
string s;
getline(cin, s);
if (cin.eof()) break;
...
}

// another version which didn't work
for (;;) {
string s;
cin >> s;
if (cin.eof()) break;
...
}
[/cpp]

In this particular problem, no whitespace was permitted except newlines.

It turns out that both these input methods, while legal on UNIX-style text files which always end lines with newlines (newline as a line terminator), fail on DOS-style text files which may omit the final newline at EOF (newline as a line separator).

The problem is with the use of istream::eof to detect the end of input.
If there is a newline terminating the final line, then eof() will not return true until the last getline or >> fails. Without a newline, eof() will return true immediately after extraction of the last line.

There are two ways to get around this:
* use the return code of getline() to determine success
* use !cin.good() as a safe substitute for cin.eof()

.good() refers to the success of the last I/O operation, so it's correct with or without a trailing newline when used with >>.

Lessons:
* avoid istream::eof(), use istream::good() instead
* it would be kind of this site to ensure that judging data follows the standard conventions for text files: one newline character terminating every line

Good luck to all, and may your WAs always be legitimate.

Regards,
Christopher

I"m also no friend of testing for EOF or whatsoever manually (actually never did it - and never needed to). Not only the return value of getline can be used, but e.g. reading int until EOF can be done the following way:

Yes, that's a great technique too.

I'd forgotten about that: istream has a truth-conversion operator (not sure if it converts to bool or to void *) which returns the same result as good().

Quite useful.