Re: A little more help ;)

[Paolo Carlini <pcarlini at unitus dot it>]

Nathan Myers wrote:

> I would not expect fstat to consider stdin to be a file.
> (It looks like a "character special file".)  
>
> Anyhow, the cin.peek() ought to call underflow(), which should do a 
> ::read(0,buf,8k), which should return 4 and put all four characters 
> into the buffer.  Then in_avail() should report them, and shouldn't 
> call showmanyc() at all, so ::fstat() should not even be called.
Thanks. I'll investigate along these lines.

> (BTW, do you know the strace command?  E.g. try "strace /bin/echo foo"
> and look at the last few lines.)
I know the command even if I don't use it so much.
In the specific case, I think that a plain gdb session should be able
to tell me what's going wrong: it's sure that showmanyc it's mistakenly
called, just I didn't (yet) put a breakpoint in the preceding cin.peek()
:(

> About the putback buffer: if a character has been put back
> in the separate putback buffer, in_avail() should return 1.
> After that character has been read out, in_avail() should 
> find the buffer empty and call showmanyc(), which may fix up
> the pointers and then report what's in the regular buffer and 
> any more characters it can discover.  In other words, the 
> putback buffer is involved in all virtuals, too.
Ok.
In fact, in underflow, a char is read from the buffer (and xsgetn
not called) only if M_pback_init is true, like this:

	  if (_M_pback_init)
	    {
	      _M_pback_destroy();
	      if (_M_in_cur < _M_in_end)
		{
		  __ret = traits_type::to_int_type(*_M_in_cur);
		  if (__bump)
		    _M_in_cur_move(1);
	  	  return __ret;
		}
	    }

Paolo.