[Bug libc/1017] New: gethostid should not sign-extend ids on 64-bit hosts

[Bug libc/1017] New: gethostid should not sign-extend ids on 64-bit hosts

[eggert at gnu dot org]

I'm following up to Tim Waugh's bug report today on bug-coreutils@gnu.org:

	The gethostid() function returns a long int, and on 64-bit platforms
	this can mean that the 32-bit value becomes sign-extended.  Here is a
	patch to only display the 32-bit value in hostid().

	Original bug report:

	  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=160078

Here's the underlying problem.  The program:

   #include <unistd.h>
   #include <stdio.h>
   int  main (void) { printf ("%lx", gethostid ()); }

prints something like "80b05874" on a 64-bit Solaris host, but
"ffffffff80b05874" on a 64-bit GNU/Linux host.

This is because glibc is incompatible with Solaris on 64-bit hosts.
glibc sign-extends the 32-bit host ID, whereas Solaris does not.

POSIX <http://www.opengroup.org/onlinepubs/009695399/functions/gethostid.html>
says merely that the host ID must be a "32-bit identifier".  One could
argue that a number in the range -2147483648...2147483647 is a "32-bit
identifier" and therefore glibc conforms, but the Solaris
interpretation is more natural, and is less likely to break existing
code as illustrated in the above-referenced bug report and the
above-mentioned example.

I'll attach a proposed patch.

-- 
           Summary: gethostid should not sign-extend ids on 64-bit hosts
           Product: glibc
           Version: 2.3.5
            Status: NEW
          Severity: normal
          Priority: P2
         Component: libc
        AssignedTo: gotom at debian dot or dot jp
        ReportedBy: eggert at gnu dot org
                CC: glibc-bugs at sources dot redhat dot com
  GCC host triplet: 64-bit linux


http://sources.redhat.com/bugzilla/show_bug.cgi?id=1017

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libc/1017] gethostid should not sign-extend ids on 64-bit hosts

[eggert at gnu dot org]

------- Additional Comments From eggert at gnu dot org  2005-06-16 22:44 -------
Created an attachment (id=519)
 --> (http://sources.redhat.com/bugzilla/attachment.cgi?id=519&action=view)
patch for gethostid 64-bit sign extension compatibility problem


-- 


http://sources.redhat.com/bugzilla/show_bug.cgi?id=1017

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libc/1017] gethostid should not sign-extend ids on 64-bit hosts

[jakub at redhat dot com]

------- Additional Comments From jakub at redhat dot com  2005-06-16 22:51 -------
I disagree.  If you don't sign extend, then gethostid () on the same system
may return different id between 32-bit and 64-bit programs (for 32-bit
program it returns some negative value, for 64-bit program 2^32 bigger value).
When hostid(1) wants to print a signed value where only low 32 bits are
supposed to have the actual id as hexadecimal value (i.e. unsigned), it should
mask out all bits but the low 32.

-- 


http://sources.redhat.com/bugzilla/show_bug.cgi?id=1017

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libc/1017] gethostid should not sign-extend ids on 64-bit hosts

[drepper at redhat dot com]

------- Additional Comments From drepper at redhat dot com  2005-09-22 17:25 -------
The code as it is now is correct.  Sign extension is necessary.  If you want to
complain, track down the person who starting using a signed value for such an
obvious unsigned use case.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |WONTFIX


http://sourceware.org/bugzilla/show_bug.cgi?id=1017

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.