Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-28 Thread Rob 


--- Igor Robul [EMAIL PROTECTED] wrote:

 Rob wrote:
 
 --- Igor Robul [EMAIL PROTECTED] wrote:
   
 
 First:
 NULL return from dlsym() does not always imply an
 error. dlsym() can return NULL because it has an
 empty list, but this does not set the error
 indicator of dlerror(), because it is not an error.
 However, if an error occurs, dlsym() has NULL as
 a return (that's probably most sensible return).
   
 
  From manual page:
 
 The dlsym() function returns a null pointer if the
 symbol cannot be 
 found, and sets an error condition
  which may be queried with dlerror().

This is a matter of symantic and logic.

I /can/ read this as follows:

If the symbol cannot be found, then dlsym() returns
a null pointer *AND* sets the error condition.

However, dlsym() can also return the null pointer,
*WITHOUT* setting the error condition. The latter
case seems to be better specified in the Linux
manpages of dlsym(). Hence the conflict between
FreeBSD and Linux/Grace, I guess.

Rob.





__ 
Yahoo! Mail - PC Magazine Editors' Choice 2005 
http://mail.yahoo.com
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]

Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-28 Thread Rob 


--- Igor Robul [EMAIL PROTECTED] wrote:

 Rob wrote:
 
 
 The dlsym() function returns a null pointer if the
 symbol cannot be 
 found, and sets an error condition
  which may be queried with dlerror().
 
 
 
 This is a matter of symantic and logic.
 
 I /can/ read this as follows:
 
 If the symbol cannot be found, then dlsym() returns
 a null pointer *AND* sets the error condition.
   
 
 Ok, English is not my native language, and I even
 had not learned it in 
 school :-).

Neither is it for me, though I had some English
language classes at high school.
However, I assume everybody to understand the
language of logic and symantics when it comes to
computer problems :).

 However, dlsym() can also return the null pointer,
 *WITHOUT* setting the error condition. The latter
 case seems to be better specified in the Linux
 manpages of dlsym(). Hence the conflict between
 FreeBSD and Linux/Grace, I guess.
   
 
 Maybe. But I see conflict in using
 previous/non-grace result from dlerror(), which
 I think grace should not do.

Indeed, so the proper way is to call dlerror()
prior to a dlsym() call, to reset any previous
error condition. Something like this:

  /* clear error condition */
  dlerror();

  /* use dlsym and check for error */
  data = dlsym();
  if ( (error = dlerror()) != NULL ) {
  errmsg(error);
  return FAILURE;
  }


Rob.




__ 
Yahoo! FareChase: Search multiple travel sites in one click.
http://farechase.yahoo.com
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]

Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-27 Thread Rob 
--- Igor Robul [EMAIL PROTECTED] wrote:

 %mkdir 
 %cd /
 %cat ~/.grace/gracerc.user
 USE pow TYPE f_of_dd FROM /usr/lib/libm.so
 %ls
 %xmgrace
 %

OK, great.

Meanwhile, I came closer to the real problem.
You don't need to create the link in /usr/X11R6/lib.

The problem is dlerror(). Whenever one of the
dl-functions is called, it will set the error
indicator in dlerror() in case of problems.

For some unclear reason, the error indicator in
dlerror is set when grace starts-up. I don't
know why yet (I'll discuss that with the grace
mailing list).

When grace uses the dl-functions in src/dlmodule.c,
which is caused by the USE... lines in
gracerc.user, something like this happens:

   dlopen(library name, MODE);
   if (dlerror() != NULL) {
  report_error();
  exit(1);
   }

Now, dlerror() is set, but not by dlopen(), but
by some other part at start-up. A workaround is
following:

   /* reset any possible earlier error */
   dlerror();

   dlopen(library name, MODE);
   if (dlerror() != NULL) {
  report_error();
  exit(1);
   }

When I patch grace with this dummy dlerror(),
prior to the dl-function calls, all works like
a charm (and you don't need the link in
/usr/X11R6/lib anymore).

As I said before, I now have to find out why
grace activates the dlerror() at start-up.

Regards,
Rob.


__
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]

Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-27 Thread Rob 

--- Igor Robul [EMAIL PROTECTED] wrote:

 Rob wrote:
 
 
 When I patch grace with this dummy dlerror(),
 prior to the dl-function calls, all works like
 a charm (and you don't need the link in
 /usr/X11R6/lib anymore).
 
 As I said before, I now have to find out why
 grace activates the dlerror() at start-up.
   
 
 So, problem is on grace side, not on FreeBSD side.
 I think there are some differences in execution
 environment for grace on Linux and FreeBSD.

Then we have to tell the grace developper, who
is a devote Linux user, about the oddities on
FreeBSD, or we have to come up with a reasonable
patch to make it work properly on FreeBSD.

Meanwhile I digged a little deeper in the grace
source code, to find out at what place exactly
the dlerror() error-indicator is set. When
grace initializes its GUI, there is this kind
of code:

---
   XtAppContext app_con;
   Display *disp = NULL;
   char *display_name = NULL;

   XtSetLanguageProc(NULL, NULL, NULL); 
   XtToolkitInitialize();
   app_con = XtCreateApplicationContext();

   disp = XOpenDisplay(display_name);
---

(I have simplified this code snippet a bit, for
this example; also, grace uses Motif for its GUI).

Before the XOpenDisplay() call, dlerror() does
not have the error-indicator set, but after
that call, it has.

Is this where Linux and FreeBSD are out-of-sync?
Or does Grace something wrong here, or is this
a problem cause by FreeBSD or Xorg?

I feel I'm getting closer and closer, but still
I have not really a good idea of what's going on
here!

Thanks,
Rob.





__ 
Yahoo! Mail - PC Magazine Editors' Choice 2005 
http://mail.yahoo.com
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]

Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-27 Thread Igor Robul 

Rob wrote:


---
  XtAppContext app_con;
  Display *disp = NULL;
  char *display_name = NULL;

  XtSetLanguageProc(NULL, NULL, NULL); 
  XtToolkitInitialize();

  app_con = XtCreateApplicationContext();

  disp = XOpenDisplay(display_name);
---

(I have simplified this code snippet a bit, for
this example; also, grace uses Motif for its GUI).

Before the XOpenDisplay() call, dlerror() does
not have the error-indicator set, but after
that call, it has.

Is this where Linux and FreeBSD are out-of-sync?
Or does Grace something wrong here, or is this
a problem cause by FreeBSD or Xorg?

 

I think, that when XOpenDisplay called, ld.so tries load some libraries 
from preconfigured paths, and

last unsuccessful attempt is recorded for dlerror().

dlopen() _does not_ reset dlerror() state on sucess, it just returns non 
NULL. So you must not check dlerror() for error condition, you need 
check return result of dlopen(), and if it is NULL, then you need use 
dlerror().


So, code in grace:

  dlopen(library name, MODE);
  if (dlerror() != NULL) {
 report_error();
 exit(1);


  }


need to be:

handle = dlopen(library name, MODE);
if (handle == NULL) {  
   report_error(dlerror());


 exit(1);

}

just because dlerror() != NULL is not indicator of error.
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]

Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-27 Thread Igor Robul 

Igor Robul wrote:

dlopen() _does not_ reset dlerror() state on sucess, it just returns 
non NULL. So you must not check dlerror() for error condition, you 
need check return result of dlopen(), and if it is NULL, then you need 
use dlerror().


So, code in grace:

  dlopen(library name, MODE);
  if (dlerror() != NULL) {
 report_error();
 exit(1);


Actual code in grace is:

newkey.data = dlsym(handle, dl_function);
   if ((error = (char *) dlerror()) != NULL) {
   errmsg(error);
   dlclose(handle);
   return RETURN_FAILURE;
   }

But I think it is needed to be something like (I dont know if they need 
error variable later):


newkey.data = dlsym(handle, dl_function);
   if (  newkey.data == NULL) {
   errmsg(error=dlerror());
   dlclose(handle);
   return RETURN_FAILURE;
   }


___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]

Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-27 Thread Igor Robul 

Sorry,
I have reread manual page for dlerror() and found that it need clear 
error state after call, but

dlerror() in src/libc/gen/dlfcn.c does not do this:

#pragma weak dlerror
const char *
dlerror(void)
{
   return sorry;
}

So error is in FreeBSD libc, if I understand this correctly. I'll do PR.

___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]

Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-27 Thread Igor Robul 

Igor Robul wrote:


Sorry,
I have reread manual page for dlerror() and found that it need clear 
error state after call, but

dlerror() in src/libc/gen/dlfcn.c does not do this:

#pragma weak dlerror
const char *
dlerror(void)
{
   return sorry;
}

So error is in FreeBSD libc, if I understand this correctly. I'll do PR.


Bad day for me :-(.

Above code is from src/libc/gen/dlfcn.c, but real dlerror for dynamicaly 
linked executables is in rtdl.c, and it works as described in  manual 
page (clears error status).


So problem is somewhere else. Sorry for false info.

___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]

Re: math/grace port: libXcursor.so.1.0 not found ?? [SOLVED]

2005-10-27 Thread Rob 


--- Igor Robul [EMAIL PROTECTED] wrote:

 Rob wrote:
 
 ---
XtAppContext app_con;
Display *disp = NULL;
char *display_name = NULL;
 
XtSetLanguageProc(NULL, NULL, NULL); 
XtToolkitInitialize();
app_con = XtCreateApplicationContext();
 
disp = XOpenDisplay(display_name);
 ---
 
 (I have simplified this code snippet a bit, for
 this example; also, grace uses Motif for its GUI).
 
 Before the XOpenDisplay() call, dlerror() does
 not have the error-indicator set, but after
 that call, it has.
   
 
 I  wrote simple program to test XOpenDisplay +
 dlopen:
 
 #include X11/Xlib.h
 #include dlfcn.h
 #include stdio.h
 
 main(int argc, char **argv)
 {
 Display *d;
 void *handle;
 void *f;
 char *err;
 
 if ((err=dlerror()) != NULL) {
 fprintf(stderr,dlerror: %s\n,
 err);
 exit(1);
 }
 
 d = XOpenDisplay(NULL);
 if (d == NULL) {
 fprintf(stderr, Error in
 XOpenDisplay\n);
 exit(1);
 }
 
 if ((err=dlerror()) != NULL) {
 fprintf(stderr, dlerror: %s\n,
err);
 exit(1);
 }
 
 XCloseDisplay(d);
 
 handle = dlopen(argv[1], RTLD_NOW);
 if (handle == NULL) {
 fprintf(stderr, dlopen: %s\n,
dlerror());
 exit(1);
 }
 
 f = dlsym(handle, argv[2]);
 if (f == NULL) {
 fprintf(stderr, dlsym: %s\n,
dlerror());
 exit(1);
 }
 
 dlclose(handle);
 }
 
 Compile it with
 gcc -o t t.c -I /usr/X11R6/include -L /usr/X11R6/lib
 -lX11
 
 Then test with:
 ./t /usr/lib/libm.so pow
 
 Then
 ./t /usr/lib/libm.so pow33
 dlsym: Undefined symbol pow33
 
 So as you can see on my system all looks good. Try
 it on your system.


First:
NULL return from dlsym() does not always imply an
error. dlsym() can return NULL because it has an
empty list, but this does not set the error
indicator of dlerror(), because it is not an error.
However, if an error occurs, dlsym() has NULL as
a return (that's probably most sensible return).


Second:
I get different output than you, from this part
of the program:
 
 d = XOpenDisplay(NULL);
 if (d == NULL) {
  fprintf(stderr, Error in XOpenDisplay\n);
  exit(1);
 }

 if ((err=(char *)dlerror()) != NULL) {
  fprintf(stderr, dlerror: %s\n, err);
  exit(1);
 }

I get here:
dlerror: Shared object nss_dns.so.1 not found,
 required by t

So apparently dlerror() is set by XOpenDisplay().

Hmmmyesterday I updated my world and kernel
to most recent 5-Stable, and after that I
recompiled all my ports. Still get this problem!

This is my /etc/make.conf:
---
CFLAGS= -O -pipe
NOPROFILE=true
NO_PF=true

CUPS_OVERWRITE_BASE=yes
NO_LPR=yes

WITHOUT_NLS=yes
WITHOUT_XFT=yes
DOC_LANG=en_US.ISO8859-1
WITH_ASPELL=yes
WITH_GTK2=yes
WITH_GIMP=yes
A4=yes

WITH_X11=yes

# added by use.perl 2005-10-26 14:20:09
PERL_VER=5.8.7
PERL_VERSION=5.8.7
---

Does that cause any problems?

Rob.


__
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]