[RFC] PR5820: two CTRL-D are not enough?!?

[Nathan Myers]

On Sun, Mar 31, 2002 at 02:18:05PM +0200, Paolo Carlini wrote:
> this is one of weirdest PR I have seen in the last months and I have no 
> idea if the current behaviour of v3 is conforming or not in this respect.
> 
> Submitter maintains that usually two CTRL-D indicates end of output, 
> whereas v3 needs /four/.
> This is the original testcase:
> 
> #include <iostream>
> 
> int main()
> {
>    int i;
>    char c;
>    
>    std::cout << "Enter an integer: ";
>    std::cin >> i;
>    if (!std::cin) {
>      std::cerr << "Unexpected read problem\n";
>      return 1;
>    }   
>    std::cout << '\n' << "std::cin.eof()=" << std::cin.eof() << '\n';
>    std::cout << "The number was: " << i << '\n';
>    std::cin >> c;
>    if (std::cin) {
>      std::cerr << "Could read " << c << " after the number!\n";
>    }   
> }
> 
> I have confirmed that this happens on linux and the following is an 
> excerpt of a trace (after a "5" I had to press CTRL-D four times):
> 
> ...
> 
> SYS_197(0, 0xbffff610, 0, 0xbffff610, 0) = 0
> ioctl(0, TCGETS, {B38400 opost isig icanon echo ...}) = 0
> mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) 
> = 0x400c6000
> read(0, "5", 4096)                      = 1
> _llseek(0, -1, 0xbffff550, SEEK_CUR)    = -1 ESPIPE (Illegal seek)
> _llseek(0, 0, 0xbffff550, SEEK_CUR)     = -1 ESPIPE (Illegal seek)
> _llseek(0, 1, 0xbffff410, SEEK_CUR)     = -1 ESPIPE (Illegal seek)
> _llseek(0, 0, 0xbffff410, SEEK_CUR)     = -1 ESPIPE (Illegal seek)
> read(0, "", 4096)                       = 0
> _llseek(0, 1, 0xbffff420, SEEK_CUR)     = -1 ESPIPE (Illegal seek)
> _llseek(0, 0, 0xbffff420, SEEK_CUR)     = -1 ESPIPE (Illegal seek)
> read(0, "", 4096)                       = 0
> _llseek(0, 1, 0xbffff420, SEEK_CUR)     = -1 ESPIPE (Illegal seek)
> _llseek(0, 0, 0xbffff420, SEEK_CUR)     = -1 ESPIPE (Illegal seek)
> read(0, "", 4096)                       = 0
> write(1, "\n", 1)                       = 1
> 
> ...
> 
> In fact, the same testcase compiled by icc is not affected by the 
> "problem"...

I agree this is a serious bug.  A single ctrl-D should suffice,
but maybe the terminal mode is eating one of them.  I don't see
a need for any calls to _llseek, never mind extra calls to read.

Nathan Myers
ncm at cantrip dot org