Ticket History #2941: krb5-1.4 is unbuildable on HPUX 10.20

[Please CC me on any replies; I'm not on any krb lists.]

I've just tried to build V5 for the first time under HPUX 10.20.

./configure blows up because it can't figure out what to do with
threading. Using --disable-thread-support lets it finish (20 minutes
later, sheesh), but then make blows up on the -very first file-
because it tried to build threads.c, which seems extremely wrong
given that I've already told it not to bother using threads at all...

You can find all the output from ./configure & make in
http://foner.www.media.mit.edu/people/foner/Sys/krb5-1.4-build-failed

What should I do now?

[I'm trying to beat a just-announced deadline in which I need to
get support working for some of these machines by 3/1 before NIS
support is discontinued, so I'd like to try to get this working
as soon as possible, even it involves ugly workarounds... Should
I be disturbed by all the tickets on krbdev being 1-8 -years- old?]

Thanks!

Date: Wed, 23 Feb 2005 19:46:16 -0500 (EST)
From: "Tom Yu via RT" <rt-comment@krbdev.mit.edu>




Try using --disable-shared --enable-static as well as
--disable-thread-support. That will prevent the initializer/finalizer
support from being built.

Did "make distclean; ./configure --disable-thread-support
--disable-shared --enable-static; make". Still no joy.
Blew up again, in a slightly different way, still in threads.c.

http://foner.www.media.mit.edu/people/foner/Sys/krb5-1.4-build-failed-1
has this attempt.

Ideas?

Date: Thu, 24 Feb 2005 01:00:28 -0500 (EST)
From: "Ken Raeburn via RT" <rt-comment@krbdev.mit.edu>

On Feb 23, 2005, at 22:10, Lenny Foner via RT wrote:

Looks like it's having trouble finding a 64-bit type. Some of the
updates in GSSAPI require 64-bit support.

Let me guess... HP-UX 10.20 has stdint.h or inttypes.h, but unlike all
of our test systems, doesn't define int64_t in either?

To save a bunch o' hits on the webserver, I just did the obvious grep
(without filtering out any of the bogus stuff) and got the following.
It sure looks like sys/_inttypes.h defines int64_t, and that
sys/types.h includes it (see second grep), but I'm not sure yet
if anything includes sys/types.h. The configure run didn't find it:

checking for uint64_t... no
checking for int64_t... no

Is there anything else I can check on this end, or is this a case of
someone who's more familiar w/the krb configure checks tweaking them?

[oob] 02:20:04 /usr/include> grep-tree int64_t
./sys/_inttypes.h:typedef long long int64_t; /* 64-bit signed integer */
./sys/_inttypes.h:typedef unsigned long long uint64_t; /* 64-bit unsigned integer */
./sys/_inttypes.h:typedef long int64_t; /* 64-bit signed integer */
./sys/_inttypes.h:typedef unsigned long uint64_t; /* 64-bit unsigned integer */
./sys/_inttypes.h:typedef int64_t intmax_t; /* largest signed integer supported */
./sys/_inttypes.h:typedef uint64_t uintmax_t; /* largest unsigned integer supported */
./sys/_inttypes.h:typedef int64_t int_least64_t;
./sys/_inttypes.h:typedef int64_t int_fast64_t;
./sys/_inttypes.h:typedef uint64_t uint_least64_t;
./sys/_inttypes.h:typedef uint64_t uint_fast64_t;
./sys/mp.h: int64_t ipsw_savearea; /* Pointer to interrupt status word */
./sys/param.h:# define MAX_LOCK64_SIZE (uint64_t)(MAX_LARGE_FILE + 1) /* 64-bit lock size */
./sys/param.h:# define MAX_LOCK_SIZE (uint64_t)(MAX_LARGE_FILE+1)
./sys/param.h: (sizeof(bytes) == sizeof(int64_t)? \
./sys/param.h: (unsigned)((uint64_t)(bytes) >> DEV_BSHIFT): \
./sys/param.h: ((uint64_t)((uint32_t)db) << DEV_BSHIFT)
./sys/pci.h:typedef void (*RD_DD)(PHOST_ADAPTER, uint64_t *, uint64_t *);
./sys/pci.h:typedef void (*WR_DD)(PHOST_ADAPTER, uint64_t *, uint64_t);
./sys/portal.h: * __type - data type (i.e. int32_t, int64_t, char, ...)
./sys/portal.h: * SET_MASK_BIT(SIGN_BIT(int64_t), int64_t);
./sys/portal.h: * __type - data type (i.e. int32_t, int64_t, char, ...)
./sys/portal.h: * __type - data type (i.e. int32_t, int64_t, char, ...)
./sys/portal.h: * __old_type - data type (i.e. int32_t, int64_t, char, ...)
./sys/portal.h: * __new_type - data type >= __old_type (i.e. int32_t, int64_t, char, ...)
./sys/portal.h: * int64_t i;
./sys/portal.h: * i = SIGN_EXTEND(c, char, int64_t);
./sys/stat.h:# define ad_long_t int64_t
./sys/stat.h:# define ino_t int64_t
./sys/stat.h:# define ad_long_t long /* 64BC-REVISIT: "int64_t"? */
./sys/statvfs.h:# define fsfilcnt_t uint64_t
./sys/statvfs.h:# define fsblkcnt_t uint64_t
./sys/statvfs.h:# define ad_ulong_t uint64_t
./sys/types.h: typedef uint64_t ino_t; /* Just API goes wide for now... */
./sys/types.h: typedef int64_t fpos64_t; /* 64bit position inside a file */
./sys/types.h: typedef int64_t fpos_t; /* 64bit position inside a file */
./sys/types.h: typedef uint64_t fsblkcnt64_t; /* blocks within a file system */
./sys/types.h: typedef uint64_t fsblkcnt_t; /* 64-bit block count */
./sys/types.h: typedef int64_t off64_t; /* 64bit offsets and sizes */
./sys/types.h: typedef int64_t off_t; /* For 64-bit offsets and sizes */
./sys/types.h: typedef uint64_t fsfilcnt64_t; /* free file nodes */
./sys/types.h: typedef uint64_t fsfilcnt_t; /* 64-bit free file nodes */
./sys/types.h: typedef int64_t blkcnt64_t; /* 64-bit # of blocks */
./sys/types.h: typedef int64_t blkcnt_t; /* 64-bit # of blocks */
./sys/types.h: typedef int64_t ssize_t; /* Signed version of size_t */
./sys/types.h: typedef uint64_t size_t; /* Type returned by sizeof() */
./sys/types.h: typedef uint64_t rlim64_t;
./sys/types.h: typedef uint64_t rlim_t;
./sys/types.h: typedef uint64_t space_t;
./sys/types.h: int64_t lbl_rp;
./sys/types.h: int64_t lbl_sp;
./sys/types.h: int64_t lbl_s[17];
./sys/types.h: int64_t lbl_ss[1];
./sys/types.h: int64_t lbl_sf[4];
./sys/types.h: typedef int64_t k_off_t;
./sys/types.h: typedef uint64_t k_rlim_t;
./sys/types.h: typedef int64_t k_blkcnt_t;
./sys/user.h: uint64_t u_ipsw; /* ipsw for single stepping */
./sys/user.h: int64_t u_gr1; /* value for general register 1 */
./sys/user.h: int64_t u_gr2; /* value for general register 2 */
./sys/vmmac.h:# define btop(x) ((sizeof(x) == sizeof(int64_t))? \
./sys/vmmac.h: (unsigned)((uint64_t)(x) >> PGSHIFT): \
./sys/vmmac.h:# define ptoo(x) ((k_off_t)((((uint64_t)((uint32_t)x)) << PGSHIFT)))
./sys/vmmac.h:# define btorp(x) ((sizeof(x) == sizeof(int64_t))? \
./sys/vmmac.h: (unsigned)(((uint64_t)(x) + NBPG -1) >> PGSHIFT): \
./inttypes.h:** uint64_t u;
./machine/save_state.h: int64_t ss_reserved;
./machine/save_state.h: int64_t ss_gr1;
./machine/save_state.h: int64_t ss_rp;
./machine/save_state.h: int64_t ss_gr3;
./machine/save_state.h: int64_t ss_gr4;
./machine/save_state.h: int64_t ss_gr5;
./machine/save_state.h: int64_t ss_gr6;
./machine/save_state.h: int64_t ss_gr7;
./machine/save_state.h: int64_t ss_gr8;
./machine/save_state.h: int64_t ss_gr9;
./machine/save_state.h: int64_t ss_gr10;
./machine/save_state.h: int64_t ss_gr11;
./machine/save_state.h: int64_t ss_gr12;
./machine/save_state.h: int64_t ss_gr13;
./machine/save_state.h: int64_t ss_gr14;
./machine/save_state.h: int64_t ss_gr15;
./machine/save_state.h: int64_t ss_gr16;
./machine/save_state.h: int64_t ss_gr17;
./machine/save_state.h: int64_t ss_gr18;
./machine/save_state.h: int64_t ss_gr19;
./machine/save_state.h: int64_t ss_gr20;
./machine/save_state.h: int64_t ss_gr21;
./machine/save_state.h: int64_t ss_gr22;
./machine/save_state.h: int64_t ss_arg3;
./machine/save_state.h: int64_t ss_arg2;
./machine/save_state.h: int64_t ss_arg1;
./machine/save_state.h: int64_t ss_arg0;
./machine/save_state.h: uint64_t ss_dp;
./machine/save_state.h: uint64_t ss_ret0;
./machine/save_state.h: uint64_t ss_ret1;
./machine/save_state.h: uint64_t ss_sp;
./machine/save_state.h: uint64_t ss_gr31;
./machine/save_state.h: uint64_t ss_cr11;
./machine/save_state.h: uint64_t ss_pcoq_head;
./machine/save_state.h: uint64_t ss_pcsq_head;
./machine/save_state.h: uint64_t ss_pcoq_tail;
./machine/save_state.h: uint64_t ss_pcsq_tail;
./machine/save_state.h: uint64_t ss_cr15;
./machine/save_state.h: uint64_t ss_cr19;
./machine/save_state.h: uint64_t ss_cr20;
./machine/save_state.h: uint64_t ss_cr21;
./machine/save_state.h: uint64_t ss_cr22;
./machine/save_state.h: uint64_t ss_cpustate;
./machine/save_state.h: uint64_t ss_sr4;
./machine/save_state.h: uint64_t ss_sr0;
./machine/save_state.h: uint64_t ss_sr1;
./machine/save_state.h: uint64_t ss_sr2;
./machine/save_state.h: uint64_t ss_sr3;
./machine/save_state.h: uint64_t ss_sr5;
./machine/save_state.h: uint64_t ss_sr6;
./machine/save_state.h: uint64_t ss_sr7;
./machine/save_state.h: uint64_t ss_cr0;
./machine/save_state.h: uint64_t ss_cr8;
./machine/save_state.h: uint64_t ss_cr9;
./machine/save_state.h: uint64_t ss_cr10;
./machine/save_state.h: uint64_t ss_cr12;
./machine/save_state.h: uint64_t ss_cr13;
./machine/save_state.h: uint64_t ss_cr24;
./machine/save_state.h: uint64_t ss_cr25;
./machine/save_state.h: uint64_t ss_cr26;
./machine/save_state.h: uint64_t ss_reserved2[3];
./regex.h: typedef int64_t ssize_t; /* Signed version of size_t */
./stdio.h: typedef int64_t fpos64_t; /* 32bit position inside a file */
./stdio.h: typedef int64_t fpos_t; /* 64bit position inside a file */

[oob] 02:20:25 /usr/include> grep-tree _inttypes.h
./sys/_inttypes.h: * @(#)_inttypes.h: $Revision: 1.2.98.8 $ $Date: 98/02/18 06:22:55 $
./sys/_inttypes.h: * @(#)_inttypes.h $Date: 98/02/18 06:22:55 $ $Revision: 1.2.98.8 $ PATCH_10.20 (PHKL_14172)
./sys/_inttypes.h:/* end of _inttypes.h */
./sys/pci.h:#include "../h/_inttypes.h"
./sys/pci.h:#include <sys/_inttypes.h>
./sys/types.h:#include "../h/_inttypes.h"
./sys/types.h:#include <sys/_inttypes.h>
./curses.h:#include <sys/_inttypes.h>
./grp.h:# include <sys/_inttypes.h>
./inttypes.h:#include "../h/_inttypes.h"
./inttypes.h:#include <sys/_inttypes.h>
./machine/save_state.h:# include "../h/_inttypes.h"
./machine/save_state.h:# include <sys/_inttypes.h>
[oob] 02:21:56 /usr/include>

From looking at the headers you've made available, it looks like
sys/_inttypes.h is always included, but it only defines the 64-bit
types in some circumstances, namely when it doesn't define
__STDC_32_MODE__; it defines that if __LP64__ is not defined (i.e., not
a 64-bit build), and __cplusplus is not defined (i.e., plain C), and
either __STDC__ is not defined (non-ANSI C build, which we're not going
to do) or __STDC_EXT__ is defined.

So I guess the question is how to get __STDC_EXT__ defined. In modern
systems there's often a header that tests some compiler-defined flags
for indicating desired standards compliance, and sets a bunch of other
macros controlling which declarations get exposed by other headers.
Looks like sys/stdsyms.h is the one for hpux 10. I don't see it doing
anything with __STDC_EXT__ except testing it, though, so this one may
be a compiler-defined symbol.

Ah... first hit in google is a discussion on the gcc-bugs list about
using __STDC_EXT__ with gcc 2.95.1, the reply claiming that the change
to gcc had been made (for some future release, maybe 3.0?) to define
__STDC_EXT__ unless the "-ansi" option is given. Lenny, what version
of gcc are you using? If it's 2.95, try

configure CPPFLAGS=-D__STDC_EXT__ --enable-static --disable-shared ...

(You might not need to disable the thread support; threads.c always
gets compiled first, but just shouldn't do anything interesting after
including some header files if the thread support is turned off, and it
was the info on shared-library support in those headers that caused the
failure the first time.)

I have a very slow HP_UX 10.20 system, so built krb5-1.4 on it.

I am using gcc-3.3.2, and built with:

CPPFLAGS=-D__hpux
../src/configure --enable-shared --enable-kdc-replay-cache \
--prefix=/krb5 --disable-thread-support

Output from configure has:
checking for unit64_t... yes
checking for int64_t... yes

It compiles *BUT* klist seg faults, and rlogin and rsh get:
error getting credentials: ASN.1 bad return from gmtime



Ken Raeburn via RT wrote:


--

Douglas E. Engert <DEEngert@anl.gov>
Argonne National Laboratory
9700 South Cass Avenue
Argonne, Illinois 60439
(630) 252-5444

Date: Thu, 24 Feb 2005 17:22:50 -0500 (EST)
From: "Tom Yu via RT" <rt-comment@krbdev.mit.edu>

It may be worth noting that the recommended invocation of the HP-UX
C compiler, 'cc -Ae', defines '-D_HPUX_SOURCE', which turns on most of
the extended namespace. By default the headers on HP-UX provide a
very restricted namespace.

Er, this compilation is being done under (a very old) gcc,
since the HP-UX "good" compiler costs money and we never
sprung for it, given gcc's existence.

It certainly doesn't look like the current compilation (under gcc) has
-D_HPUX_SOURCE being used, which come to think of it -is- somewhat
weird; many other things I've compiled using configure have tended to
set that, so that could explain something. (Not -everything-, mind
you, but many things. stunnel & emacs, for example, don't, so it's
not clear that everything really needs this.)

Just for yucks, I tried adding it manually to the giant list of others
in the threads.c compilation and retrying just that one invocation of
gcc, and it blows up identically.

Btw, I built a guest account for Ken to poke around with, though I'm
not sure what his current state is; if anyone else would like one, let
me know.

On Feb 24, 2005, at 17:46, Lenny Foner via RT wrote:

Yeah, unfortunately the default reply-to address is the "file a comment
and send mail to subscribers of the bug list but not to the person who
reported the bug" address. I've been editing the headers in my replies
to use the "including the reporter of the bug" address...

Ken

Update on status:

With the two dependent tickets (gmtime_r, deltat parsing) worked around,
with static libraries only, with thread support disabled, with gcc 2.7.2
and "configure ... CPPFLAGS=-D__STDC_EXT__", it builds, and passes all
the non-dejagnu tests.

Then Lenny installed dejagnu, and I found more problems. :-)
Several of the ftp, rcp, and rsh tests fail.
Some kadmin, changepw problems too.