10:50 a.m.
Justin Vallon writes:
| On Mon, 31 May 1999, Marcus Thiessel wrote:
(usually use when the process waits very very short time on
|
> IO). Signals don't get delivered to a process who is currently in that
| > state. If the process is blocked on that particular IO you cannot
| > SIGKILL it.
|
| Also, NFS seems to cause this (as well). Locking especially.
NFS is a special guest in this case. NFS is designed to never get
killed by anybody but to wait. You know the Hitchhiker's Guide (was
it)? Remember the story where the spaceship didn't leave before the
planet died because no tissues for the passengers were delivered to
the ship. So the automatic ship controls (nfsd) decide not to
leave. In fact the control program knew that every life on planet
would die, but decided to wait because if you wait long enough the
lifeforms will return and some small day tissues will be available
again. ;-)
| > Maybe force --with-dynamic to be yes.
|
| Hmm. Even without --with-dynamic, [tx]emacs are dynamically linked with
| libXm, libXt, libX11, libc, etc.
|
| --dynamic=no still gave me an executable dynamically linked with libXm,
| libXt, libX11, libc, etc.
HP-UX 10.20 and --with-dynamic don't work together as expected. I had
soem problems with --with-dynamic={yes,no} when building XEmacs on
9.05.
| Is src/s/hpux10.h correct here?
|
I skiped HP-UX 10.20 almost completely with regards to XEmacs beta
testing. But as far as I remember there were some build reports:
http://www.xemacs.org/list-archives/xemacs-build-reports/9901/msg00024.html
This was the last one from the mail archives (xemacs-build-reports)
and with no problems reported.
| Which again would be shared, but ld chokes on the -Wl,-E (which would be
| appropriate if cc -Wl,-E which passes -E to linker stage).
I never managed to build XEmacs on HP-UX 9.x with HP C compiler. The
one we had at our department was unusable because nobody cared about
it. I always used gcc with HP-UX 9.05. So I can tell.
| One (unrelated) thing I did notice between hpux9-shr.h and hpux10-shr.h.
| hpux10-shr.h:
|
| /* We must turn off -g since it forces -static. */
| #ifdef __GNUC__
| #undef C_DEBUG_SWITCH
| #define C_DEBUG_SWITCH
| #endif
|
| But hpux9-shr.h has that disabled:
|
| #if 0 /* No longer needed, since in current GCC -g no longer does that.
| */
| /* We must turn off -g since it forces -static. */
| #ifdef __GNUC__
| #undef C_DEBUG_SWITCH
| #define C_DEBUG_SWITCH
| #endif
| #endif
|
| Maybe the hpux10-shr.h also be #if 0'ed? Same in hpux11-shr.h.
For sure these file could be cleaned up a little, I guess.
--Marcus
5:50 a.m.
New subject: [21.1.2] Problems running dumped xemacs on hppa1.1-hp-hpux10.20
Solved. /net + my home dir name /a/vallon is a problem.
On Tue, 1 Jun 1999, Marcus Thiessel wrote:
Justin Vallon writes:
| On Mon, 31 May 1999, Marcus Thiessel wrote:
| > (usually use when the process waits very very short time on
| > IO). Signals don't get delivered to a process who is currently in that
| > state. If the process is blocked on that particular IO you cannot
| > SIGKILL it.
|
| Also, NFS seems to cause this (as well). Locking especially.
Or, trying to access /net/unknown-host with a broken automounter, as it
turns out.
The universe had aligned against me:
1) Somebody (FreeBSD) maps the temp-mount prefix /a/HOST to /net/HOST.
2) My home directory is /a/vallon.
3) (Our) HP automounter and/or /net is severely broken at times.
xemacs decided that /a/vallon was really better off being accessed as
/net/vallon (ha). The broken automounter caused a hard-hang looking for
host vallon. Eventually, my kill -9's got through (days later!).
Here's my start at a fix (with gratuitous context):
--- lisp/paths.el 1999-06-02 15:21:37-04 1.1
+++ lisp/paths.el 1999-06-02 16:42:39-04
＠＠ -257,26 +257,27 ＠＠
;; Our code handles the following empirically observed conventions:
;; /net is an actual directory! (some systems are not broken!)
;; /net/HOST -> /tmp_mnt/net/HOST (`standard' old Sun automounter)
;; /net/HOST -> /tmp_mnt/HOST (BSDI 4.0)
;; /net/HOST -> /a/HOST (Freebsd 2.2.x)
;; /net/HOST -> /amd/HOST (seen in amd sample config files)
;;
;; If your system has a different convention, you may have to change this.
;; Don't forget to send in a patch!
(when (file-directory-p "/net")
(append
(when (file-directory-p "/tmp_mnt")
(if (file-directory-p "/tmp_mnt/net")
'(("\\`/tmp_mnt/net/" . "/net/"))
'(("\\`/tmp_mnt/" . "/net/"))))
- (when (file-directory-p "/a")
+ ;; /a, but not hp
+ (when (and (file-directory-p "/a") (not (file-exists-p "/bin/hp")))
'(("\\`/a/" . "/net/")))
(when (file-directory-p "/amd")
'(("\\`/amd/" . "/net/")))
)))
))
(if (running-temacs-p)
(initialize-xemacs-paths))
It seems a bit agressive to assume presense of /net and /a implies /a gets
mapped to /net. Even with a working /net directory, I would be hosed, I
believe, when my home directory /a/vallon gets morphed into /net/vallon.
[ Could somebody tell the FreeBSD people to pick a better automount
directory name? 1/2 :-). ]
The [ ! -f /bin/hp ] is bogus, but it seems that something should be
checked here. It might be rare, but /a is a bad name anyway. Is there a
more xemacs-ey (memq '(freebsd) xemacs-build-arch) that could be used?
Here's another place:
--- lib-src/update-elc.sh 1999-06-02 15:48:46-04 1.1
+++ lib-src/update-elc.sh 1999-06-02 15:50:39-04
＠＠ -65,10 +65,12 ＠＠
# Account for various system automounter configurations
if test -d "/net"; then
if test -d "/tmp_mnt/net"; then tdir="/tmp_mnt/net"; else
tdir="/tmp_mnt"; fi
- EMACS_DIR=`echo "$EMACS_DIR" | \
- sed -e "s|^${tdir}/|/net/|" -e "s|^/a/|/net/|" -e
"s|^/amd/|/net/|"`
- CANON_PWD=`echo "$CANON_PWD" | \
- sed -e "s|^${tdir}/|/net/|" -e "s|^/a/|/net/|" -e
"s|^/amd/|/net/|"`
+ EMACS_DIR=`echo "$EMACS_DIR" | sed -e "s|^${tdir}/|/net/|" -e
"s|^/amd/|/net/|"`
+ CANON_PWD=`echo "$CANON_PWD" | sed -e "s|^${tdir}/|/net/|" -e
"s|^/amd/|/net/|"`
+ if [ ! -f /bin/hp ] ; then
+ EMACS_DIR=`echo "$EMACS_DIR" | sed -e "s|^/a/|/net/|"`
+ CANON_PWD=`echo "$CANON_PWD" | sed -e "s|^/a/|/net/|"`
+ fi
fi
REAL="$EMACS_DIR/`basename $EMACS`"
Again, only do the translation if [ ! -f /bin/hp ].
Is there a need for this script to get a canonical directory name? It
looks like it runs batch-remove-old-elc, and batch-byte-compile. Is `pwd`
stored anyplace permanently? Is a canonical dir name required?
After these changes, xemacs worked! I'll probably send a build success
once I get some auxiliary packages built.
-Justin
vallon(a)mindspring.com
9306
days inactive
9308
days old
1 comments
2 participants
participants (2)
-
Justin Vallon -
Marcus Thiessel