https://bugzilla.novell.com/show_bug.cgi?id=866669 https://bugzilla.novell.com/show_bug.cgi?id=866669#c1 Bernhard Wiedemann changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |bwiedemann@suse.com AssignedTo|bnc-team-screening@forge.pr |schwab@suse.com |ovo.novell.com | --- Comment #1 from Bernhard Wiedemann 2014-03-17 15:53:21 CET --- I can reproduce it here on 12.3 following the docs on http://docs.oracle.com/cd/E19683-01/816-0214/6m6nf1par/index.html #include main() { XDR *xdrs; //u_quad_t ull; u_longlong_t ull; xdr_u_longlong_t(xdrs, &ull); } this is part of glibc-devel -> assigning to maintainer -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=866669 https://bugzilla.novell.com/show_bug.cgi?id=866669#c3 --- Comment #3 from Bernhard Wiedemann 2014-03-17 16:41:12 CET --- err. indeed example was wrong. should be #include main() { XDR *xdrs; unsigned long long ull; xdr_u_longlong_t(xdrs, &ull); } -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=866669 https://bugzilla.novell.com/show_bug.cgi?id=866669#c5 --- Comment #5 from Michael Matz 2014-03-18 12:17:34 UTC --- (In reply to comment #4)

Where is u_longlong_t defined?

I believe that's exactly the problem. It's defined nowhere in linux, and xdr_u_longlong_t takes an u_quad_t in glibc. Solaris' variant takes an u_longlong_t (and presumably defines it in rpc/types.h, can't check). The problem of course is, that the name of the function suggests it takes an argument that's compatible with "unsigned long long *" (which is what Solaris' version indeed seems to do), but in fact doesn't. On 32bit machines it will also take long long * in linux, but on 64bit machine it just takes a long * (because quad_t is defined to some 64bit type, so long is enough on 64bit). So, for Solaris compatibility it would be better to change the prototype to take an u_longlong_t (and define it to unsigned long long for all platforms). But it could change behaviour of existing programs that expect u_quad_t to work (which wouldn't be compatible with u_longlong_t if it were defined as above on 64bit machines). I don't know a good solution :-/ -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=866669 https://bugzilla.novell.com/show_bug.cgi?id=866669#c7 --- Comment #7 from Andreas Schwab 2014-03-18 14:00:17 CET --- The manual says that xdr_u_longlong_t is identical to xdr_hyper, and the latter takes a quad_t as well on glibc. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=866669 https://bugzilla.novell.com/show_bug.cgi?id=866669#c9 --- Comment #9 from Thorsten Kukuk 2014-03-18 13:15:43 UTC --- I did only apply a patch somebody else posted, as written in that ChangeLog entry. The interface in this form is coming from SUN, more specific from the TI-RPC sources SUN released as open source at that time. Don't know about Solaris, but the ti-rpc sources and other BSD variants still have the same prototype today: extern bool_t xdr_u_longlong_t(XDR *, u_quad_t *); The question is, if the definition of u_quad_t is is correct. From what I could find u_quad_t should be defined as u_int64 on x86-64. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=866669 https://bugzilla.novell.com/show_bug.cgi?id=866669#c11 --- Comment #11 from Thorsten Kukuk 2014-03-18 16:47:34 UTC --- Ok, this reminds me on one important thing: All types of xdr_*() functions need to have the same size on any system, else a 32bit system for example couldn't communicate with a 64bit system any longer. Thus u_quad_t/u_int64_t for xdr_u_longlong_t on x86-64 seems to be correct. You only need to remind when this functions where written and that no 64bit systems did exist at that time. I propose to not change anything to not break all existing RPC software out in the world. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=866669 https://bugzilla.novell.com/show_bug.cgi?id=866669#c13 Andreas Schwab changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |RESOLVED Resolution| |INVALID --- Comment #13 from Andreas Schwab 2014-03-20 09:22:00 CET --- Since glibc matches what BSD does this is not a bug. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.