Rb_class_new_instance behaves strange

ruby-talk
1

My code is behaving different, when im doing this change:

(a)

VALUE self = rb_class_new_instance(a.n, a.argv, klass);

to:

(b)

VALUE self = rb_funcall2(klass, rb_intern(“new”), a.n, a.argv);

What is the difference ???

Problem: with (a) then we cannot inheirit from a SWIG class.
Solution: with (b) then inheiritance works fine.

Why does this make a difference ???

···

–
Simon Strandgaard

2

Hi,

My code is behaving different, when im doing this change:

(a)

VALUE self = rb_class_new_instance(a.n, a.argv, klass);

to:

(b)

VALUE self = rb_funcall2(klass, rb_intern(“new”), a.n, a.argv);

What is the difference ???

(a) passes the given block but (b) doesn’t, in normal case.

Problem: with (a) then we cannot inheirit from a SWIG class.
Solution: with (b) then inheiritance works fine.

How do they work fine or don’t?
And which version of ruby?

···

At Tue, 15 Apr 2003 20:02:56 +0900, Simon Strandgaard wrote:

–
Nobu Nakada

3

My code is behaving different, when im doing this change:

(a)

VALUE self = rb_class_new_instance(a.n, a.argv, klass);

to:

(b)

VALUE self = rb_funcall2(klass, rb_intern(“new”), a.n, a.argv);

What is the difference ???

(a) passes the given block but (b) doesn’t, in normal case.

Problem: with (a) then we cannot inheirit from a SWIG class.
Solution: with (b) then inheiritance works fine.

How do they work fine or don’t?

OK. This is what im trying to do.
note: “Embed::Redirect” is a SWIG class.

cat test.rb
class RubyView < Embed::Redirect
def initialize
super
end
def insert
repaint
end
end

If I try instantiate this “RubyView” class with (a), then it doesn’t work
and Im getting a error from ruby (sorry its converted into a c++exception).

EXCEPTION (RUBY):
name=cannot invoke ruby-function
where=./a.rb:6
class=TypeError
message=wrong argument type RubyView (expected Data)
backtrace= from ./a.rb:6:in repaint' from ./a.rb:6:in insert’

If I try instantiate with (b) then everything works (no errors).

And which version of ruby?

ruby -v
ruby 1.8.0 (2003-03-03) [i386-freebsd5.0]

···

On Tue, 15 Apr 2003 21:16:55 +0900, nobu.nokad wrote:

At Tue, 15 Apr 2003 20:02:56 +0900, Simon Strandgaard wrote:

–
Simon Strandgaard

4

And which version of ruby?

ruby -v
ruby 1.8.0 (2003-03-03) [i386-freebsd5.0]

BTW this is the SWIG im using :slight_smile:

swig -version
SWIG Version 1.3.18
Copyright (c) 1995-1998
University of Utah and the Regents of the University of California
Copyright (c) 1998-2002
University of Chicago
Compiled with g++

Please see http://www.swig.org for reporting bugs and further information

···

On Tue, 15 Apr 2003 14:39:03 +0200, Simon Strandgaard wrote:

On Tue, 15 Apr 2003 21:16:55 +0900, nobu.nokad wrote:

–
Simon Strandgaard

5

BTW this is the SWIG im using :slight_smile:

Can you give the code generated by SWIG, if it's not too long or send me
it in private email

Guy Decoux

6

fetch this tar.gz file:
http://metaeditor.sourceforge.net/embed/rubyembed-0.1.tar.gz

the SWIG generated file is
rubyembed-0.1.tar.gz/source/redirect_wrap.cpp

···

On Tue, 15 Apr 2003 23:11:24 +0900, ts wrote:

BTW this is the SWIG im using :slight_smile:

Can you give the code generated by SWIG, if it’s not too long or send me
it in private email

Guy Decoux

–
Simon Strandgaard

7

http://metaeditor.sourceforge.net/embed/rubyembed-0.1.tar.gz

I've not found 1.3.18, at sourceforge there is only .19, .17

pigeon% swig -version

SWIG Version 1.3.19
Copyright (c) 1995-1998
University of Utah and the Regents of the University of California
Copyright (c) 1998-2002
University of Chicago
Compiled with g++

Please see http://www.swig.org for reporting bugs and further information
pigeon%

pigeon% ruby -v
ruby 1.8.0 (2003-04-10) [i686-linux]
pigeon%

pigeon% tail -18 rubycpp.cpp
        VALUE self = rb_class_new_instance(a.n, a.argv, klass);
        //VALUE self = rb_funcall2(klass, rb_intern("new"), a.n, a.argv);
        return self;
}

VALUE New(string klass) {
        NewArguments arg(klass.c_str(), 0, 0);
        int error = 0;
        VALUE self = rb_protect(NewWrap, reinterpret_cast<VALUE>(&arg), &error);
        if(error) {
                std::ostringstream o;
                o << "error creating " << klass;
                throw RUBY_CPP::RubyError::Create(o.str());
        }
        return self;
}

} // end of namespace RUBY_CPP
pigeon%

and I've no problems

pigeon% ./testswig
main: enter
test.rb: enter
test.rb: leave
aeditorlib: started (no-swig)
-- Test1 simple ----------------------
view.ctor: enter
redirect.ctor: hello
rubyview.initialize: overloaded
objects += 401f530c
view.ctor: leave
view.insert: enter
rubyview.insert: enter
redirect.repaint: enter
viewqt.repaint: bingo
redirect.repaint: leave
rubyview.insert: leave
view.insert: leave
view.dtor: enter
objects -= 401f530c
view.dtor: leave
-- Test2 multiple instances ----------------------
[1]
view.ctor: enter
redirect.ctor: hello
rubyview.initialize: overloaded
objects += 401f4858
view.ctor: leave
view.insert: enter
rubyview.insert: enter
redirect.repaint: enter
viewqt.repaint: bingo
redirect.repaint: leave
rubyview.insert: leave
view.insert: leave
[2]
view.ctor: enter
redirect.ctor: hello
rubyview.initialize: overloaded
objects += 401f4808
view.ctor: leave
view.insert: enter
rubyview.insert: enter
redirect.repaint: enter
viewqt.repaint: bingo
redirect.repaint: leave
rubyview.insert: leave
view.insert: leave
[3]
view.insert: enter
rubyview.insert: enter
redirect.repaint: enter
viewqt.repaint: bingo
redirect.repaint: leave
rubyview.insert: leave
view.insert: leave
[4]
view.ctor: enter
redirect.ctor: hello
rubyview.initialize: overloaded
objects += 401f47b8
view.ctor: leave
view.insert: enter
rubyview.insert: enter
redirect.repaint: enter
viewqt.repaint: bingo
redirect.repaint: leave
rubyview.insert: leave
view.insert: leave
[5]
view.dtor: enter
objects -= 401f47b8
view.dtor: leave
view.dtor: enter
objects -= 401f4808
view.dtor: leave
view.dtor: enter
objects -= 401f4858
view.dtor: leave
-- End ------------------------------------
redirect.dtor: hello
redirect.dtor: hello
redirect.dtor: hello
redirect.dtor: hello
aeditorlib: stopped
main: leave (0)
pigeon%

Guy Decoux

8

Hi,

···

At Tue, 15 Apr 2003 22:44:39 +0900, Simon Strandgaard wrote:

BTW this is the SWIG im using :slight_smile:

Can you give the code generated by SWIG, if it’s not too long or send me
it in private email

Guy Decoux

fetch this tar.gz file:
http://metaeditor.sourceforge.net/embed/rubyembed-0.1.tar.gz

the SWIG generated file is
rubyembed-0.1.tar.gz/source/redirect_wrap.cpp

SWIG doesn’t seem to use rb_define_alloc_func(), so allocator
of Object is used by rb_class_new_instance(). “new” works fine
because SWIG redefines Redirect.new (but it is deprecated now).

–
Nobu Nakada

9

http://metaeditor.sourceforge.net/embed/rubyembed-0.1.tar.gz

I have just installed the newest SWIG version 1.3.19 and generated a new
wrapper. But it still FAILs!!

./testswig
main: enter
test.rb: enter
test.rb: leave
aeditorlib: started (with-swig)
– Test1 simple ----------------------
view.ctor: enter
rubyview.initialize: overloaded
redirect.ctor: hello
objects += 8108cf0
view.ctor: leave
view.insert: enter
rubyview.insert: enter
view.dtor: enter
objects -= 8108cf0
view.dtor: leave
EXCEPTION (RUBY):
name=cannot invoke ruby-function
where=./test.rb:17
class=TypeError
message=wrong argument type RubyView (expected Data)
backtrace= from ./test.rb:17:in repaint' from ./test.rb:17:in insert’

main: leave (1)

tail -18 rubycpp.cpp
VALUE self = rb_class_new_instance(a.n, a.argv, klass);
//VALUE self = rb_funcall2(klass, rb_intern(“new”), a.n, a.argv);
return self;
}

VALUE New(string klass) {
NewArguments arg(klass.c_str(), 0, 0);
int error = 0;
VALUE self = rb_protect(NewWrap, reinterpret_cast(&arg), &error);
if(error) {
std::ostringstream o;
o << "error creating " << klass;
throw RUBY_CPP::RubyError::Create(o.str());
}
return self;
}

} // end of namespace RUBY_CPP

swig -version

SWIG Version 1.3.19
Copyright (c) 1995-1998
University of Utah and the Regents of the University of California
Copyright (c) 1998-2002
University of Chicago
Compiled with g++

Please see http://www.swig.org for reporting bugs and further information

ruby -v
ruby 1.8.0 (2003-03-03) [i386-freebsd5.0]

and I’ve no problems

There is 2 binaries, I think you executed the wrong one.

pigeon% ./testswig
main: enter
test.rb: enter
test.rb: leave
aeditorlib: started (no-swig)
^^^^^^^
^^^^^^^
If you ran the “testswig” program, it should say “with-swig”…
I think you ran the “test” program :slight_smile:

Try again :slight_smile:

···

On Wed, 16 Apr 2003 00:34:33 +0900, ts wrote:

–
Simon Strandgaard

10

Is this a problem SWIG ?
Or is this a problem with RUBY ?

···

On Wed, 16 Apr 2003 01:32:33 +0900, nobu.nokad wrote:

Hi,

At Tue, 15 Apr 2003 22:44:39 +0900, > Simon Strandgaard wrote:

BTW this is the SWIG im using :slight_smile:

Can you give the code generated by SWIG, if it’s not too long or send me
it in private email

Guy Decoux

fetch this tar.gz file:
http://metaeditor.sourceforge.net/embed/rubyembed-0.1.tar.gz

the SWIG generated file is
rubyembed-0.1.tar.gz/source/redirect_wrap.cpp

SWIG doesn’t seem to use rb_define_alloc_func(), so allocator
of Object is used by rb_class_new_instance(). “new” works fine
because SWIG redefines Redirect.new (but it is deprecated now).

–
Simon Strandgaard

11

Hi,

···

At Wed, 16 Apr 2003 01:25:33 +0900, Simon Strandgaard wrote:

SWIG doesn’t seem to use rb_define_alloc_func(), so allocator
of Object is used by rb_class_new_instance(). “new” works fine
because SWIG redefines Redirect.new (but it is deprecated now).

Is this a problem SWIG ?

SWIG (1.3.19 at least) seems to not support ruby 1.8.

–
Nobu Nakada

12

[Nobu said:]

SWIG doesn’t seem to use rb_define_alloc_func(), so allocator
of Object is used by rb_class_new_instance(). “new” works fine
because SWIG redefines Redirect.new (but it is deprecated now).

[And then Simon responded:]

Is this a problem SWIG ?
Or is this a problem with RUBY ?

[And now Lyle says:]

So far, I’ve purposely avoided making any changes to support Ruby’s new
allocation scheme since (at least in its initial implementation in Ruby
1.7.x) there seemed to be some problems. There’s also the issue of
backwards compatibility with Ruby versions 1.6.x and earlier. If SWIG is
to generate code that will compile under both Ruby 1.6 and Ruby 1.8,
there will presumably need to be a compile-time check of the Ruby
version number.

Since, as Nobu notes, redefining ‘new’ works for both Ruby 1.6 and 1.8,
I’ve left it that way until now. If Matz, Nobu and others feel confident
about the new allocation implementation in 1.8 I can revisit the issue.

13

1.7.x) there seemed to be some problems. There’s also the issue of
backwards compatibility with Ruby versions 1.6.x and earlier. If SWIG is
to generate code that will compile under both Ruby 1.6 and Ruby 1.8,
there will presumably need to be a compile-time check of the Ruby
version number.

Nothing new for this, from bdb

#if RUBY_VERSION_CODE < 180
#define StringValue(x,y) rb_str2cstr(x,y)
#define StringValuePtr(x) STR2CSTR(x)
#define SafeStringValue(x) Check_SafeStr(x)
#endif

#if RUBY_VERSION_CODE >= 180
rb_define_alloc_func(bdb_cEnv, bdb_env_s_alloc);
#else
rb_define_singleton_method(bdb_cEnv, “allocate”, bdb_env_s_alloc, 0);
#endif

etc, etc

Guy Decoux

14

Hi,

···

In message “Re: rb_class_new_instance behaves strange” on 03/04/16, Lyle Johnson lyle@users.sourceforge.net writes:

Since, as Nobu notes, redefining ‘new’ works for both Ruby 1.6 and 1.8,
I’ve left it that way until now. If Matz, Nobu and others feel confident
about the new allocation implementation in 1.8 I can revisit the issue.

I’m confident. Tell me if you see any problem that I don’t know.

						matz.
15

Hi,

So far, I’ve purposely avoided making any changes to support Ruby’s
new allocation scheme since (at least in its initial implementation in
Ruby 1.7.x) there seemed to be some problems. There’s also the issue
of backwards compatibility with Ruby versions 1.6.x and earlier. If
SWIG is to generate code that will compile under both Ruby 1.6 and
Ruby 1.8, there will presumably need to be a compile-time check of the
Ruby version number.

1.8 defines HAVE_RB_DEFINE_ALLOC_FUNC for that purpose.

— Source/Modules/ruby.cxx~ Tue Mar 18 01:52:37 2003
+++ Source/Modules/ruby.cxx Wed Apr 16 02:03:09 2003
@@ -479,6 +479,11 @@ public:
break;
case CONSTRUCTOR_ALLOCATE:

  •  Printv(s, "#ifdef HAVE_RB_DEFINE_ALLOC_FUNC\n", NIL);

  •  Printv(s, tab4, "rb_define_alloc_func(", klass->vname,

  •    ", ", wname, ");\n", NIL);

  •  Printv(s, "#else\n", NIL);
 Printv(s, tab4, "rb_define_singleton_method(", klass->vname,
   ", \"new\", ", wname, ", -1);\n", NIL);

  •  Printv(s, "#endif\n", NIL);
 Replaceall(klass->init,"$allocator", s);
 break;

@@ -830,5 +835,7 @@ public:
need_result = 1;
Printf(f->code, “VALUE vresult = SWIG_NewClassInstance(self, SWIGTYPE%s);\n”, Char(SwigType_manglestr(t)));

  •  Printf(f->code, "#ifndef HAVE_RB_DEFINE_ALLOC_FUNC\n");
 Printf(f->code, "rb_obj_call_init(vresult, argc, argv);\n");

  •  Printf(f->code, "#endif\n");

    } else if (current == CONSTRUCTOR_INITIALIZE) {
    need_result = 1;
    @@ -1329,6 +1336,8 @@ public:

    String *s = NewString(“”);

  • Printv(s, “#ifndef HAVE_RB_DEFINE_ALLOC_FUNC\n”, NIL);
    Printv(s, tab4, “rb_undef_method(CLASS_OF(”, klass->vname,
    “), "new");\n”, NIL);

  • Printv(s, “#endif\n”, NIL);
    Replaceall(klass->init,“$allocator”, s);
    Replaceall(klass->init,“$initializer”, “”);

···

At Wed, 16 Apr 2003 01:45:49 +0900, Lyle Johnson wrote:

–
Nobu Nakada

16

nobu.nokada@softhome.net wrote in message news:200304151635.h3FGZNHQ025574@sharui.nakada.kanuma.tochigi.jp…

Hi,

SWIG doesn’t seem to use rb_define_alloc_func(), so allocator
of Object is used by rb_class_new_instance(). “new” works fine
because SWIG redefines Redirect.new (but it is deprecated now).

Is this a problem SWIG ?

SWIG (1.3.19 at least) seems to not support ruby 1.8.

And the practical upshot is?
Mind you, at least you guy’s got to look at simon’s tutorial. What do
you think. Any comments for us humble users :slight_smile:

···

At Wed, 16 Apr 2003 01:25:33 +0900, > Simon Strandgaard wrote:

17

ts wrote:

Nothing new for this, from bdb

Thanks very much, Guy; I’ve incorporated this into the SWIG CVS and it
should be available for SWIG 1.3.20.

18

Thanks very much for the patches, Nobu; I’ve incorporated them into the
SWIG CVS and they should be available in SWIG 1.3.20.

One aspect of this change is still unclear to me: what is now the
correct way to indicate that an extension class is an abstract class?
Previously, SWIG was either redefining ‘new’ for concrete extension
classes, or undefining ‘new’ for abstract extension classes.

···

nobu.nokada@softhome.net wrote:

1.8 defines HAVE_RB_DEFINE_ALLOC_FUNC for that purpose.

19

Hi,

One aspect of this change is still unclear to me: what is now the
correct way to indicate that an extension class is an abstract class?
Previously, SWIG was either redefining ‘new’ for concrete extension
classes, or undefining ‘new’ for abstract extension classes.

Sorry, I missed it. You can make them abstract by undefining
allocator with rb_undef_alloc_func().

Please replace all STR2CSTR to StringValue, and apply this
patch.

— Lib/ruby/rubyhead.swg~ Wed Feb 26 04:22:36 2003
+++ Lib/ruby/rubyhead.swg Wed Apr 16 06:11:34 2003
@@ -76,3 +76,26 @@ typedef struct {
void (*destroy)(void *);
} swig_class;

···

At Wed, 16 Apr 2003 04:47:13 +0900, Lyle Johnson wrote:
+
+/* Don’t use for expressions have side effect /
+#ifndef RB_STRING_VALUE
+#define RB_STRING_VALUE(s) (TYPE(s) == T_STRING ? (s) : ((volatile VALUE *)&(s) = rb_str_to_str(s)))
+#endif
+#ifndef StringValue
+#define StringValue(s) RB_STRING_VALUE(s)
+#endif
+#ifndef StringValuePtr
+#define StringValuePtr(s) RSTRING(RB_STRING_VALUE(s))->ptr
+#endif
+#ifndef StringValueLen
+#define StringValueLen(s) RSTRING(RB_STRING_VALUE(s))->len
+#endif
+#define SafeStringValue(v) do {\

  • StringValue(v);\
  • rb_check_safe_str(v);
    +} while (0)

+#ifndef HAVE_RB_DEFINE_ALLOC_FUNC
+#define rb_define_alloc_func(klass, func) rb_define_singleton_method(klass, “new”, func, -1)
+#define rb_undefine_alloc_func(klass) rb_undef_method(CLASS_OF(klass), “new”)
+#endif
— Source/Modules/ruby.cxx~ Tue Mar 18 01:52:37 2003
+++ Source/Modules/ruby.cxx Wed Apr 16 06:11:48 2003
@@ -479,6 +479,6 @@ public:
break;
case CONSTRUCTOR_ALLOCATE:

  •  Printv(s, tab4, "rb_define_singleton_method(", klass->vname,

  •    ", \"new\", ", wname, ", -1);\n", NIL);

  •  Printv(s, tab4, "rb_define_alloc_func(", klass->vname,

  •    ", ", wname, ");\n", NIL);
 Replaceall(klass->init,"$allocator", s);
 break;

@@ -830,5 +830,7 @@ public:
need_result = 1;
Printf(f->code, “VALUE vresult = SWIG_NewClassInstance(self, SWIGTYPE%s);\n”, Char(SwigType_manglestr(t)));

  •  Printf(f->code, "#ifndef HAVE_RB_DEFINE_ALLOC_FUNC\n");
 Printf(f->code, "rb_obj_call_init(vresult, argc, argv);\n");

  •  Printf(f->code, "#endif\n");

    } else if (current == CONSTRUCTOR_INITIALIZE) {
    need_result = 1;
    @@ -1329,6 +1331,5 @@ public:

    String *s = NewString(“”);

  • Printv(s, tab4, “rb_undef_method(CLASS_OF(”, klass->vname,
  •  "), \"new\");\n", NIL);
  • Printv(s, tab4, “rb_undef_alloc_func(”, klass->vname, “);\n”, NIL);
    Replaceall(klass->init,“$allocator”, s);
    Replaceall(klass->init,“$initializer”, “”);

–
Nobu Nakada

20

Sorry, I missed it. You can make them abstract by undefining
allocator with rb_undef_alloc_func().

OK.

Please replace all STR2CSTR to StringValue, and apply this
patch.

Don’t you mean replace all STR2CSTR with StringValuePtr? I think
StringValue() returns a String object (a VALUE) while StringValuePtr()
returns the underlying C character array.

···

nobu.nokada@softhome.net wrote: