Re: libstdc++/4686: Output disappearing after operator<<( streambuf* )

[Carlo Wood <carlo@alinoe.com>]

The following reply was made to PR libstdc++/4686; it has been noted by GNATS.

From: Carlo Wood <carlo@alinoe.com>
To: kanze@gabi-soft.de
Cc: gcc-gnats@gcc.gnu.org
Subject: Re: libstdc++/4686: Output disappearing after operator<<( streambuf* )
Date: Fri, 26 Oct 2001 06:00:57 +0200

 On Thu, Oct 25, 2001 at 05:49:35PM -0000, kanze@gabi-soft.de wrote:
 > #include <sstream>
 > #include <ostream>
 > #include <iostream>
 > 
 > int
 > main()
 > {
 >     std::istringstream  i1( "3\n" ) ;
 >     std::cout << i1.rdbuf() ;
 >     std::cout << "4\n" ;
 >     return 0 ;
 > }
 
 The reason that nothing is written the second time is beause badbit is set
 for std::cout.  It seems to be set here:
 
 0x0804e134 in std::ostream::sentry::~sentry() (__in_chrg=0xbffff510) at /usr/local/gcc-cvs-3.0/include/g++-v3/bits/std_ostream.h:200
 200                   _M_os.setstate(ios_base::badbit);
 (gdb) l
 195             // XXX MT
 196             if (_M_os.flags() & ios_base::unitbuf && !uncaught_exception())
 197               {
 198                 // Can't call flush directly or else will get into recursive lock.
 199                 if (_M_os.rdbuf() && _M_os.rdbuf()->pubsync() == -1)
 200                   _M_os.setstate(ios_base::badbit);
 201               }
 202           }
 
 right after the call to std::cout << i1.rdbuf();
 
 This indeed seems to be a bug, because - this code seems to try to flush
 and sets badbit when that fails.  However, there is nothing to flush because
 std::cout is both 1) unbuffered, 2) empty.  That is hardly a case where
 an *attempt* to flush it can cause an error.
 
 -- 
 Carlo Wood <carlo@alinoe.com>