Forums » Ruby-core » ruby_script ?

Hi list,

That's my first post on this list, so please be patient with me. :-)

The code below leads me on a SEGV in tcltklib.so

$ cat test.c
#include <ruby.h>

int main()
{
 ruby_init();
 ruby_script("test");
 ruby_init_loadpath();
 rb_eval_string("require 'tk'");
 return 0;
}
$ gcc -o test test.c -L/opt/ruby-1.8.4/lib -lruby
-I/opt/ruby-1.8.4/lib/ruby/1.8/i686-linux
$ ./test
/opt/ruby-1.8.4/lib/ruby/site_ruby/1.8/i686-linux/tcltklib.so: [BUG]
Segmentation fault
ruby 1.8.4 (2005-12-24) [i686-linux]

zsh: abort      ./test
$

After debugging it, I figure out that the bug comes from the fact
rb_argv0 was not initialized (the SEGV happens on line 7932 of the
file tcltklib.c). Then, I fixed my progam this way:

> $ cat test.c
> #include <ruby.h>
>

EXTERN VALUE rb_progname;
EXTERN VALUE rb_argv0;

> int main()
> {
>   ruby_init();
>   ruby_script("test");

 rb_argv0 = rb_progname;

>   ruby_init_loadpath();
>   rb_eval_string("require 'tk'");
>   return 0;
> }

I noticed that using ruby_options also solves the problem since
ruby_process_options does a "rb_argv0 = rb_progname" right after
ruby_script, as I do in my code. Using ruby_options leads my program
to sleep since ruby_options reads stdin which I don't want. Thus, I'm
wondering whether ruby_script should always initialized rb_argv0 to
avoid such issue.

Best regards,

>>>>> "N" == =?ISO-8859-1?Q?Nicolas Despr=E8s?= <ISO-8859-1> writes:

N> The code below leads me on a SEGV in tcltklib.so

 The bug is in tk, not in ruby

Guy Decoux

On 3/25/06, ts <decoux@moulon.inra.fr> wrote:
> >>>>> "N" == =?ISO-8859-1?Q?Nicolas Despr=E8s?= <ISO-8859-1> writes:
>
> N> The code below leads me on a SEGV in tcltklib.so
>
>  The bug is in tk, not in ruby
>

It may not be so obvious. Maybe tk is right to assume that rb_argv0 is
initialized before it gets required. In that case the bug is in Ruby.
If tk is wrong to assume that, then it should maybe uses rb_progname
instead of rb_argv0.

Cheers,

>>>>> "N" == =?ISO-8859-1?Q?Nicolas Despr=E8s?= <ISO-8859-1> writes:

N> It may not be so obvious. Maybe tk is right to assume that rb_argv0 
is
N> initialized before it gets required. In that case the bug is in Ruby.
N> If tk is wrong to assume that, then it should maybe uses rb_progname
N> instead of rb_argv0.

 Well tk must at least call StringValuePtr() rather than trying to 
access
 directly RSTRING(value)->ptr


Guy Decoux

Hi,

In message "Re: [BUG] ruby_script ?"
    on Sun, 26 Mar 2006 01:18:27 +0900, ts <decoux@moulon.inra.fr> 
writes:

| Well tk must at least call StringValuePtr() rather than trying to access
| directly RSTRING(value)->ptr

The attached are modifications for potential (but no sure) bugs.  Note
that it's not tested at all.

Index: tcltklib.c
===================================================================
RCS file: /var/cvs/src/ruby/ext/tk/tcltklib.c,v
retrieving revision 1.44
diff -p -u -1 -r1.44 tcltklib.c
--- tcltklib.c	6 Dec 2005 16:24:08 -0000	1.44
+++ tcltklib.c	25 Mar 2006 17:12:10 -0000
@@ -1142,4 +1142,7 @@ set_max_block_time(self, time)
     default:
-        rb_raise(rb_eArgError, "invalid value for time: '%s'",
-                 RSTRING(rb_funcall(time, ID_inspect, 0, 0))->ptr);
+        {
+	    VALUE tmp = rb_funcall(time, ID_inspect, 0, 0);
+	    rb_raise(rb_eArgError, "invalid value for time: '%s'",
+		     StringValuePtr(tmp));
+	}
     }
@@ -2134,2 +2137,3 @@ ip_set_exc_message(interp, exc)
     msg = rb_funcall(exc, ID_message, 0, 0);
+    StringValue(msg);

@@ -4912,3 +4916,3 @@ ip_create_slave_core(interp, argc, argv)

-    slave->ip = Tcl_CreateSlave(master->ip, RSTRING(name)->ptr, safe);
+    slave->ip = Tcl_CreateSlave(master->ip, StringValuePtr(name), 
safe);
     if (slave->ip == NULL) {
@@ -6194,2 +6198,3 @@ lib_fromUTF8_core(ip_obj, src, encodenam

+	    StringValue(str);
             s = 
Tcl_GetByteArrayFromObj(Tcl_NewStringObj(RSTRING(str)->ptr,
@@ -6373,3 +6378,3 @@ lib_set_system_encoding(self, enc_name)
     if (Tcl_SetSystemEncoding((Tcl_Interp *)NULL,
-                              RSTRING(enc_name)->ptr) != TCL_OK) {
+                              StringValuePtr(enc_name)) != TCL_OK) {
         rb_raise(rb_eArgError, "unknown encoding name '%s'",
@@ -7931,3 +7936,3 @@ Init_tcltklib()

-    ret = ruby_open_tcl_dll(RSTRING(rb_argv0)->ptr);
+    ret = ruby_open_tcl_dll(rb_argv0 ? RSTRING(rb_argv0)->ptr : 0);
     switch(ret) {

From: Yukihiro Matsumoto <matz@ruby-lang.org>
Subject: Re: [BUG] ruby_script ?
Date: Sun, 26 Mar 2006 02:14:34 +0900
Message-ID: <1143306862.695945.12581.nullmailer@x31.priv.netlab.jp>
> The attached are modifications for potential (but no sure) bugs.  Note
> that it's not tested at all.

Thank you for your patch. But now, I have no time to test it.
If it can fix the problem, please commit it.

Hi,

In message "Re: [BUG] ruby_script ?"
    on Mon, 27 Mar 2006 11:58:44 +0900, Hidetoshi NAGAI 
<nagai@ai.kyutech.ac.jp> writes:

|Thank you for your patch. But now, I have no time to test it.
|If it can fix the problem, please commit it. 

Only after someone confirms that it works.  I don't have an
environment to compile Ruby/Tk.

							matz.

>>>>> "Y" == Yukihiro Matsumoto <matz@ruby-lang.org> writes:

Y> Only after someone confirms that it works.  I don't have an
Y> environment to compile Ruby/Tk.

 It's 1.8.4

moulon% cat a.c
#include <ruby.h>

int main()
{
    ruby_init();
    ruby_script("test");
    ruby_init_loadpath();
    rb_eval_string("require 'tk'");
    return 0;
}
moulon%

moulon% gcc -fPIC -g -O2  -I. 
-I/home/msys/decoux/local/r184/lib/ruby/1.8/i686-linux a.c 
-L/home/msys/decoux/local/r184/lib -lpthread -ldl -lcrypt -lm 
-lruby-static -rdynamic
moulon%

moulon% ./a.out
moulon%


Guy Decoux