libstdc++/9521: Garbled input from wcin

[peturr02@ru.is]

>Number:         9521
>Category:       libstdc++
>Synopsis:       Garbled input from wcin
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Jan 31 12:16:01 UTC 2003
>Closed-Date:
>Last-Modified:
>Originator:     peturr02@ru.is
>Release:        gcc-3.2.1
>Organization:
>Environment:
Red Hat Linux 8.0
>Description:
If wcin.rdbuf()->sgetc() is called followed by wcin.rdbuf()->sbumpc(), the return value is not always the same, unless the charset in use happens to be ISO-8859-1.

Reason: basic_filebuf<wchar_t>::_M_underflow_common returns wide characters to the FILE* using ungetc. This happens to work if the charset is ISO-8859-1 (because the values are the same as in UCS-4), but breaks for all other single-byte charsets.

Note that there is an even more serious bug here when dealing with multibyte charsets: in general a single wide character can have been converted from more than one narrow character. This is however shadowed by other problems in _M_underflow_common.
>How-To-Repeat:
See attachment.
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted:
----gnatsweb-attachment----
Content-Type: text/plain; name="ungetcbug.cc"
Content-Disposition: inline; filename="ungetcbug.cc"

#include <iostream>
#include <fstream>
#include <locale>
#include <cwchar>

#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>

int main()
{
	using namespace std;

	const char* name = "tmp";

	filebuf fbuf1;
	fbuf1.open(name, ios_base::out | ios_base::trunc);
	for (int i = 1; i < 256; ++i)
		fbuf1.sputc(static_cast<unsigned char>(i));
	fbuf1.close();

	int fd = open(name, O_RDONLY);
	assert(fd != -1);
	dup2(fd, 0);
	
	locale loc ("en_US.ISO-8859-15");
	locale::global(loc);
	wcin.imbue(loc);

	for (int j = 1; j < 256; ++j)
	{
		wint_t c1 = wcin.rdbuf()->sgetc();
		wint_t c2 = wcin.rdbuf()->sbumpc();
		assert(c1 == c2);
		assert(c1 != WEOF);
	}

	return 0;
}