Home / Browse / luabind / Mail / Archive

luabind

Email Archive: luabind-user (read-only)

Re: [luabind] [ANN] luabind 0.9-rc1

From: Daniel Wallin <daniel@bo...> - 2009-12-10 21:27

liam mail wrote:
> Is there a packaging mistake for the 0.9rc1 zip file, I still see the
> directory inside named luabind-0.8.1 and the speed it almost the same?

Uh, no. Are you sure you didn't just download the wrong file? I just
double checked.

-- 
Daniel Wallin
BoostPro Computing
http://www.boostpro.com

Thread View

Thread	Author	Date
[luabind] [ANN] luabind 0.9-rc1	Daniel Wallin <daniel@bo...>

Re: [luabind] [ANN] luabind 0.9-rc1

From: liam mail <liam.list@go...> - 2009-12-10 13:27

Attachments: Message as HTML

2009/12/10 Daniel Wallin <daniel@...>

> I'm happy to announce the first RC of luabind 0.9. There have been a lot
> of improvements since 0.8.1. Among the more interesting ones are the
> automatic handling of down casts, as well as better smart pointer
> support, and of course the improved performance.
>
> Many thanks to Athena Capital Research who funded a large part of the
> development for this release.
>
> Please test and report problems!
>
> Source packages available for download at:
>
>  http://sourceforge.net/projects/luabind/files/
>
>
> User visible changes since 0.8.1
> --------------------------------
>
>  * Major optimizations. For example, calling member functions should be
>   close to twice as fast.
>
>  * Can now automatically handle conversion of any pointer like type
>   that implements a get_pointer() overload.
>
>  * Now handles virtual and multiple inheritance properly.
>
>  * Automatically perform down-casts as well as up-casts when converting
>   arguments from Lua to C++.
>
>  * When converting from C++ to Lua, polymorphic instances are
>   automatically cast to their most derived type.
>
>  * Memory optimizations.
>
>  * Added `table<>` object wrapper. Can be used to limit a function
>   parameter to tables.
>
>   Previously:
>
>       void f(object const& x)
>       {
>           // assert that x is table
>       }
>
>   Now:
>
>       void f(table<object> const& x)
>       {
>       }
>
>  * Uses standard luaL_ref() and luaL_unref() functions. This should
>   simplify integration with other libraries that make use of the
>   registry.
>
>  * Deprecated the `raw()` converter policy. lua_State* arguments are now
>   automatically handled by the library.
>
>  * Fixed common problem where objects would reference a garbage
>   collected coroutine.
>
> Breaking changes
> ----------------
>
>  * Dropped support for using a custom type_info.
>
>  * `get_const_holder()' was removed, and by extension the automatic
>   smart_ptr<X> -> smart_ptr<X const>.
>
>
> Changes since 0.8.1:
> --------------------
>
> Daniel Wallin (100):
>  Remove a lot of unused code.
>  Remove unnecessary dependency on Boost.Python.
>  Don't use boost::function.
>  Remove a bunch of dead code and unnecessary includes.
>  Use mpl::vector instead of mpl::list in class_<>.
>  Simplify parameter deduction code in class_<> template.
>  Use partial specialization for is_bases<>.
>  Move "bases<>" definition to class.hpp.
>  Add missing <stdexcept> include.
>  Replace ref/unref with luaL_ref/luaL_unref.
>  Use luaL_ref/luaL_unref in weak_ref.
>  Add type_info wrapper class.
>  Add table<> object wrapper.
>  Fix placeholder macro for GCC4.0.
>  Replace include with forward declaration.
>  New instance holders. Can hold instance by any pointer type.
>  New instance storage implementation. Cleaner and faster.
>  Add default converter for boost::shared_ptr<>.
>  Use luaL_ref/luaL_unref instead of old ref() mechanism.
>  Don't try to return void* from Lua in test.
>  Add missing LUABIND_API to luabind::open().
>  Add test for creating instances in Lua threads.
>  Add luabind::get_main_thread() API.
>  Create instance table reference in the main thread.
>  New recursive overload resolver.
>  Compute argument indices with converters.
>  Automatically convert lua_State* arguments.
>  Test user defined converter that consumes 2 parameters.
>  Inline more calls when invoking functions. Improves code on GCC.
>  Cache conversion result from the scoring stage.
>  Reduce the number of API calls for numeric converters.
>  Use lua_rawgeti() instead of lua_pushnumber, lua_rawget pair.
>  Make adopt_policy::match() stateful.
>  Make const_pointer_converter stateful.
>  Make shared_ptr_converter stateful.
>  Fix builtin converter test. All converters need both conversion stages.
>  Fix tests. Need to contain subconverters.
>  Add missing detail/ref.hpp include.
>  New inheritance graph code.
>  Add make_instance() helper function.
>  Add tests for new inheritance code.
>  Add missing <luabind/wrapper_base.hpp> include.
>  Use add_cast() for wrappers as well.
>  Handle null pointer in shared_ptr_converter.
>  Rework local_id allocation in class_id_map.
>  Fix bug in id allocation.
>  Make shared_ptr_converter public.
>  Test object identity with shared_ptr_converter.
>  Handle NULL smart pointers.
>  Remove use of conversion_storage.
>  Test that adopt() leaves a weak pointer in the Lua object.
>  Hold a weak pointer in instance_holder when ownership is released.
>  Use false : x : 0 idiom to avoid evaluating x.
>  Suppress unused variable warning.
>  Keep overload chain alive in function_object.
>  Remove old unused table ref member from object_rep.
>  Store dependency refs directly in the registry.
>  Update version.hpp for v0.9.
>  Silence uninitialized variable warning.
>  Cache conversion result in value_converter.
>  Update version for 0.9 in Jamroot.
>  Add test for extending exported classes in Lua.
>  Associate polymorphic wrapper back reference with the main thread.
>  Remove some stale code.
>  Handle returning references to smart pointers.
>  Allocate storage for instances with object_rep::allocate().
>  Add missing LUABIND_API to invoke_context.
>  Use boost::scoped_ptr<> for pimpl.
>  Add missing LUABIND_API to class_id_map.
>  Fix bug where local class ids would become incorrect.
>  Supress int->bool warnings on MSVC.
>  Add missing LUABIND_BUILDING define to inheritance.cpp.
>  Use std::size_t instead of int for dependency count.
>  Const holder conversion isn't supported any more. Don't test it.
>  Use lua_tointeger() instead of casting the result of lua_tonumber().
>  Suppress a bunch of "unreferenced parameter" warnings.
>  Suppress MSVC alignment warning (C4121).
>  Suppress MSVC unreachable code warning.
>  Suppress MSVC "assignment operator could not be generated" warning.
>  Make object_rep noncopyable.
>  Cast to lua_Integer and lua_Number explicitly in converter.
>  Remove compute_arity(). This isn't used any more.
>  Revert 5134297. This caused overload ambiguities.
>  Fix leak in "super" upvalue.
>  Store main thread lua_State* in shared_ptr_deleter.
>  Fix class_info() calling lua_typename() with stack index.
>  Kill old detail::is_class_object() function.
>  Fix set_instance_value() bug that overwrote class values.
>  Don't test smart ptr features that have been removed.
>  docs: Rewrite section on smart pointers.
>  docs: Remove note about library being in beta phase.
>  Use lua_pushliteral() where possible.
>  Don't use lua_tointeger() for unsigned int on MSVC.
>  Suppress int->bool warnings on MSVC.
>  Don't leak cast-graph objects.
>  docs: Add note about functions returning smart pointers.
>  Accept "nil" in char const* converter.
>  Always use lua_tonumber() for unsigned int/long.
>  cast_graph::cast() should return the score as int.
>  Fix placeholder definitions to match changes in Boost 1.39.
>
> Evan Wies (3):
>  Don't crash if class_info() is called on non-luabind objects.
>  Add Lua function class_names() giving a list of registered types.
>  Test that class_info() don't crash when passed non-luabind objects.
>
> Kenny Peng (1):
>  Address a name conflict with termios.h.
>
> Nigel Atkinson (1):
>  Don't use same name for parameter and type.
>
> --
> Daniel Wallin
> BoostPro Computing
> http://www.boostpro.com
>
>
>
> ------------------------------------------------------------------------------
> Return on Information:
> Google Enterprise Search pays you back
> Get the facts.
> http://p.sf.net/sfu/google-dev2dev
> _______________________________________________
> luabind-user mailing list
> luabind-user@...
> https://lists.sourceforge.net/lists/listinfo/luabind-user
>
Is there a packaging mistake for the 0.9rc1 zip file, I still see the
directory inside named luabind-0.8.1 and the speed it almost the same?
Liam

Re: [luabind] [ANN] luabind 0.9-rc1

From: liam mail <liam.list@go...> - 2009-12-10 14:21

Attachments: Message as HTML

2009/12/10 liam mail <liam.list@...>

>
>
> 2009/12/10 Daniel Wallin <daniel@...>
>
> I'm happy to announce the first RC of luabind 0.9. There have been a lot
>> of improvements since 0.8.1. Among the more interesting ones are the
>> automatic handling of down casts, as well as better smart pointer
>> support, and of course the improved performance.
>>
>> Many thanks to Athena Capital Research who funded a large part of the
>> development for this release.
>>
>> Please test and report problems!
>>
>> Source packages available for download at:
>>
>>  http://sourceforge.net/projects/luabind/files/
>>
>>
>> User visible changes since 0.8.1
>> --------------------------------
>>
>>  * Major optimizations. For example, calling member functions should be
>>   close to twice as fast.
>>
>>  * Can now automatically handle conversion of any pointer like type
>>   that implements a get_pointer() overload.
>>
>>  * Now handles virtual and multiple inheritance properly.
>>
>>  * Automatically perform down-casts as well as up-casts when converting
>>   arguments from Lua to C++.
>>
>>  * When converting from C++ to Lua, polymorphic instances are
>>   automatically cast to their most derived type.
>>
>>  * Memory optimizations.
>>
>>  * Added `table<>` object wrapper. Can be used to limit a function
>>   parameter to tables.
>>
>>   Previously:
>>
>>       void f(object const& x)
>>       {
>>           // assert that x is table
>>       }
>>
>>   Now:
>>
>>       void f(table<object> const& x)
>>       {
>>       }
>>
>>  * Uses standard luaL_ref() and luaL_unref() functions. This should
>>   simplify integration with other libraries that make use of the
>>   registry.
>>
>>  * Deprecated the `raw()` converter policy. lua_State* arguments are now
>>   automatically handled by the library.
>>
>>  * Fixed common problem where objects would reference a garbage
>>   collected coroutine.
>>
>> Breaking changes
>> ----------------
>>
>>  * Dropped support for using a custom type_info.
>>
>>  * `get_const_holder()' was removed, and by extension the automatic
>>   smart_ptr<X> -> smart_ptr<X const>.
>>
>>
>> Changes since 0.8.1:
>> --------------------
>>
>> Daniel Wallin (100):
>>  Remove a lot of unused code.
>>  Remove unnecessary dependency on Boost.Python.
>>  Don't use boost::function.
>>  Remove a bunch of dead code and unnecessary includes.
>>  Use mpl::vector instead of mpl::list in class_<>.
>>  Simplify parameter deduction code in class_<> template.
>>  Use partial specialization for is_bases<>.
>>  Move "bases<>" definition to class.hpp.
>>  Add missing <stdexcept> include.
>>  Replace ref/unref with luaL_ref/luaL_unref.
>>  Use luaL_ref/luaL_unref in weak_ref.
>>  Add type_info wrapper class.
>>  Add table<> object wrapper.
>>  Fix placeholder macro for GCC4.0.
>>  Replace include with forward declaration.
>>  New instance holders. Can hold instance by any pointer type.
>>  New instance storage implementation. Cleaner and faster.
>>  Add default converter for boost::shared_ptr<>.
>>  Use luaL_ref/luaL_unref instead of old ref() mechanism.
>>  Don't try to return void* from Lua in test.
>>  Add missing LUABIND_API to luabind::open().
>>  Add test for creating instances in Lua threads.
>>  Add luabind::get_main_thread() API.
>>  Create instance table reference in the main thread.
>>  New recursive overload resolver.
>>  Compute argument indices with converters.
>>  Automatically convert lua_State* arguments.
>>  Test user defined converter that consumes 2 parameters.
>>  Inline more calls when invoking functions. Improves code on GCC.
>>  Cache conversion result from the scoring stage.
>>  Reduce the number of API calls for numeric converters.
>>  Use lua_rawgeti() instead of lua_pushnumber, lua_rawget pair.
>>  Make adopt_policy::match() stateful.
>>  Make const_pointer_converter stateful.
>>  Make shared_ptr_converter stateful.
>>  Fix builtin converter test. All converters need both conversion stages.
>>  Fix tests. Need to contain subconverters.
>>  Add missing detail/ref.hpp include.
>>  New inheritance graph code.
>>  Add make_instance() helper function.
>>  Add tests for new inheritance code.
>>  Add missing <luabind/wrapper_base.hpp> include.
>>  Use add_cast() for wrappers as well.
>>  Handle null pointer in shared_ptr_converter.
>>  Rework local_id allocation in class_id_map.
>>  Fix bug in id allocation.
>>  Make shared_ptr_converter public.
>>  Test object identity with shared_ptr_converter.
>>  Handle NULL smart pointers.
>>  Remove use of conversion_storage.
>>  Test that adopt() leaves a weak pointer in the Lua object.
>>  Hold a weak pointer in instance_holder when ownership is released.
>>  Use false : x : 0 idiom to avoid evaluating x.
>>  Suppress unused variable warning.
>>  Keep overload chain alive in function_object.
>>  Remove old unused table ref member from object_rep.
>>  Store dependency refs directly in the registry.
>>  Update version.hpp for v0.9.
>>  Silence uninitialized variable warning.
>>  Cache conversion result in value_converter.
>>  Update version for 0.9 in Jamroot.
>>  Add test for extending exported classes in Lua.
>>  Associate polymorphic wrapper back reference with the main thread.
>>  Remove some stale code.
>>  Handle returning references to smart pointers.
>>  Allocate storage for instances with object_rep::allocate().
>>  Add missing LUABIND_API to invoke_context.
>>  Use boost::scoped_ptr<> for pimpl.
>>  Add missing LUABIND_API to class_id_map.
>>  Fix bug where local class ids would become incorrect.
>>  Supress int->bool warnings on MSVC.
>>  Add missing LUABIND_BUILDING define to inheritance.cpp.
>>  Use std::size_t instead of int for dependency count.
>>  Const holder conversion isn't supported any more. Don't test it.
>>  Use lua_tointeger() instead of casting the result of lua_tonumber().
>>  Suppress a bunch of "unreferenced parameter" warnings.
>>  Suppress MSVC alignment warning (C4121).
>>  Suppress MSVC unreachable code warning.
>>  Suppress MSVC "assignment operator could not be generated" warning.
>>  Make object_rep noncopyable.
>>  Cast to lua_Integer and lua_Number explicitly in converter.
>>  Remove compute_arity(). This isn't used any more.
>>  Revert 5134297. This caused overload ambiguities.
>>  Fix leak in "super" upvalue.
>>  Store main thread lua_State* in shared_ptr_deleter.
>>  Fix class_info() calling lua_typename() with stack index.
>>  Kill old detail::is_class_object() function.
>>  Fix set_instance_value() bug that overwrote class values.
>>  Don't test smart ptr features that have been removed.
>>  docs: Rewrite section on smart pointers.
>>  docs: Remove note about library being in beta phase.
>>  Use lua_pushliteral() where possible.
>>  Don't use lua_tointeger() for unsigned int on MSVC.
>>  Suppress int->bool warnings on MSVC.
>>  Don't leak cast-graph objects.
>>  docs: Add note about functions returning smart pointers.
>>  Accept "nil" in char const* converter.
>>  Always use lua_tonumber() for unsigned int/long.
>>  cast_graph::cast() should return the score as int.
>>  Fix placeholder definitions to match changes in Boost 1.39.
>>
>> Evan Wies (3):
>>  Don't crash if class_info() is called on non-luabind objects.
>>  Add Lua function class_names() giving a list of registered types.
>>  Test that class_info() don't crash when passed non-luabind objects.
>>
>> Kenny Peng (1):
>>  Address a name conflict with termios.h.
>>
>> Nigel Atkinson (1):
>>  Don't use same name for parameter and type.
>>
>> --
>> Daniel Wallin
>> BoostPro Computing
>> http://www.boostpro.com
>>
>>
>>
>> ------------------------------------------------------------------------------
>> Return on Information:
>> Google Enterprise Search pays you back
>> Get the facts.
>> http://p.sf.net/sfu/google-dev2dev
>> _______________________________________________
>> luabind-user mailing list
>> luabind-user@...
>> https://lists.sourceforge.net/lists/listinfo/luabind-user
>>
> Is there a packaging mistake for the 0.9rc1 zip file, I still see the
> directory inside named luabind-0.8.1 and the speed it almost the same?
> Liam
>

So I tried the tar.gz packaged download and "simply dropping the source
files in a Visual Studio project" yet I get errors. find_best_match.hpp
seems to have been removed form luabind/detail
find_best_match.cpp
.\src\find_best_match.cpp(31) : error C2039: 'find_best_match' : is not a
member of 'luabind::detail'
.\src\find_best_match.cpp(33) : error C2061: syntax error : identifier
'overload_rep_base'
.\src\find_best_match.cpp(44) : error C2065: 'num_overloads' : undeclared
identifier
.\src\find_best_match.cpp(46) : error C2065: 'start' : undeclared identifier
.\src\find_best_match.cpp(46) : error C2227: left of '->match' must point to
class/struct/union/generic type
        type is ''unknown-type''
.\src\find_best_match.cpp(46) : error C2065: 'num_params' : undeclared
identifier
.\src\find_best_match.cpp(48) : error C2065: 'start' : undeclared identifier
.\src\find_best_match.cpp(48) : error C2065: 'orep_size' : undeclared
identifier
.\src\find_best_match.cpp(51) : error C2065: 'min_match' : undeclared
identifier
.\src\find_best_match.cpp(54) : error C2065: 'match_index' : undeclared
identifier
.\src\find_best_match.cpp(55) : error C2065: 'min_match' : undeclared
identifier
.\src\find_best_match.cpp(56) : error C2065: 'min_match' : undeclared
identifier
.\src\find_best_match.cpp(64) : error C2065: 'ambiguous' : undeclared
identifier
.\src\find_best_match.cpp(64) : error C2065: 'min_match' : undeclared
identifier
.\src\find_best_match.cpp(65) : error C2065: 'min_match' : undeclared
identifier
.\src\find_best_match.cpp(69) : error C2039: 'find_exact_match' : is not a
member of 'luabind::detail'
.\src\find_best_match.cpp(71) : error C2061: syntax error : identifier
'overload_rep_base'
.\src\find_best_match.cpp(78) : error C2065: 'num_overloads' : undeclared
identifier
.\src\find_best_match.cpp(80) : error C2065: 'start' : undeclared identifier
.\src\find_best_match.cpp(80) : error C2227: left of '->match' must point to
class/struct/union/generic type
        type is ''unknown-type''
.\src\find_best_match.cpp(80) : error C2065: 'num_params' : undeclared
identifier
.\src\find_best_match.cpp(81) : error C2065: 'cmp_match' : undeclared
identifier
.\src\find_best_match.cpp(81) : error C2065: 'dest' : undeclared identifier
.\src\find_best_match.cpp(81) : error C2228: left of '.push_back' must have
class/struct/union
        type is ''unknown-type''
.\src\find_best_match.cpp(81) : error C2065: 'start' : undeclared identifier
.\src\find_best_match.cpp(82) : error C2065: 'start' : undeclared identifier
.\src\find_best_match.cpp(82) : error C2065: 'orep_size' : undeclared
identifier

Re: [luabind] [ANN] luabind 0.9-rc1

From: Tony Kostanjsek <lobotony666@ya...> - 2009-12-10 15:12

>
> So I tried the tar.gz packaged download and "simply dropping the  
> source files in a Visual Studio project" yet I get errors.  
> find_best_match.hpp seems to have been removed form luabind/detail
> find_best_match.cpp
> .\src\find_best_match.cpp(31) : error C2039: 'find_best_match' : is  
> not a member of 'luabind::detail'
> .\src\find_best_match.cpp(33) : error C2061: syntax error :  
> identifier 'overload_rep_base'
> .\src\find_best_match.cpp(44) : error C2065: 'num_overloads' :  
> undeclared identifier


I encountered the same error, but could build and run my project after  
removing find_best_match.cpp. That was 0.9 from github, 1 or 2 weeks  
ago, haven't tried the newest version yet.

Re: [luabind] [ANN] luabind 0.9-rc1

From: liam mail <liam.list@go...> - 2009-12-10 16:01

Attachments: Message as HTML

2009/12/10 Tony Kostanjsek <lobotony666@...>

> >
> > So I tried the tar.gz packaged download and "simply dropping the
> > source files in a Visual Studio project" yet I get errors.
> > find_best_match.hpp seems to have been removed form luabind/detail
> > find_best_match.cpp
> > .\src\find_best_match.cpp(31) : error C2039: 'find_best_match' : is
> > not a member of 'luabind::detail'
> > .\src\find_best_match.cpp(33) : error C2061: syntax error :
> > identifier 'overload_rep_base'
> > .\src\find_best_match.cpp(44) : error C2065: 'num_overloads' :
> > undeclared identifier
>
>
> I encountered the same error, but could build and run my project after
> removing find_best_match.cpp. That was 0.9 from github, 1 or 2 weeks
> ago, haven't tried the newest version yet.
>
>
> ------------------------------------------------------------------------------
> Return on Information:
> Google Enterprise Search pays you back
> Get the facts.
> http://p.sf.net/sfu/google-dev2dev
> _______________________________________________
> luabind-user mailing list
> luabind-user@...
> https://lists.sourceforge.net/lists/listinfo/luabind-user
>
Thanks Tony this that is what I did in the end.
Daniel would you mind sharing your speed tests which you have previously
claimed were on par with Swig as I am not seeing the speeds you are talking
about.
In this releases notes you say "Major optimizations. For example, calling
member functions should be close to twice as fast."
Luabind 9.0rc1 tests http://pastebin.com/f72880a97
Luabind 8.1 tests http://code.google.com/p/oolua/wiki/Speed_comparisons

Now I know Luabind can do more than OOLua and I would like to have results
which are fair, so if there is anything which you think I should be doing in
the tests or more defines to speed it up please do inform me.
Thanks
Liam

Re: [luabind] [ANN] luabind 0.9-rc1

From: Daniel Wallin <daniel@bo...> - 2009-12-10 18:49

liam mail wrote:
> Daniel would you mind sharing your speed tests which you have previously
> claimed were on par with Swig as I am not seeing the speeds you are talking
> about.

I will, but my benchmark suite currently isn't usable outside my
development environment. I will try to make time to fix it up, and when
I have I will publish the code.

> In this releases notes you say "Major optimizations. For example, calling
> member functions should be close to twice as fast."
> Luabind 9.0rc1 tests http://pastebin.com/f72880a97
> Luabind 8.1 tests http://code.google.com/p/oolua/wiki/Speed_comparisons
> 
> Now I know Luabind can do more than OOLua and I would like to have results
> which are fair, so if there is anything which you think I should be doing in
> the tests or more defines to speed it up please do inform me.

Which compiler are you using for this? I just tried you benchmark, after
some fiddling to make it build, and while my results match yours for
0.8.1, for 0.9 they are completely different:

> OOLua access (average elapsed time):	0.698
> OOLua passing derived to a function that wants a base  (average elapsed time):	0.719
> OOLua passing derived with multiple bases to a function that wants a base  (average elapsed time):	0.747
> OOLua virtual function  (average elapsed time):	0.338
> OOLua pure virtual function  (average elapsed time):	0.341
> OOLua - Lua function which takes a base and calls a virtual function (average elapsed time): 0.309
> Swig access (average elapsed time):	1.582
> Swig passing derived to a function that wants a base  (average elapsed time):	0.797
> Swig passing derived with multiple bases to a function that wants a base  (average elapsed time):	0.835
> Swig virtual function  (average elapsed time):	0.747
> Swig pure virtual function  (average elapsed time):	0.755
> Swig - Lua function which takes a base and calls a virtual function (average elapsed time): 0.74
> Luabind access (average elapsed time):	1.089
> Luabind passing derived to a function that wants a base  (average elapsed time):	0.659
> Luabind passing derived with multiple bases to a function that wants a base (average elapsed time):	0.673
> Luabind virtual function  (average elapsed time):	0.515
> Luabind pure virtual function  (average elapsed time):	0.517
> Luabind - Lua function which takes a base and calls a virtual function (average elapsed time): 0.502

This is with:

  gcc version 4.3.3 (Ubuntu 4.3.3-5ubuntu4)

-- 
Daniel Wallin
BoostPro Computing
http://www.boostpro.com

Re: [luabind] [ANN] luabind 0.9-rc1

From: liam mail <liam.list@go...> - 2009-12-10 19:37

Attachments: Message as HTML

2009/12/10 Daniel Wallin <daniel@...>

> liam mail wrote:
> > Daniel would you mind sharing your speed tests which you have previously
> > claimed were on par with Swig as I am not seeing the speeds you are
> talking
> > about.
>
> I will, but my benchmark suite currently isn't usable outside my
> development environment. I will try to make time to fix it up, and when
> I have I will publish the code.
>
> Thanks that would be good.


> > In this releases notes you say "Major optimizations. For example, calling
> > member functions should be close to twice as fast."
> > Luabind 9.0rc1 tests http://pastebin.com/f72880a97
> > Luabind 8.1 tests http://code.google.com/p/oolua/wiki/Speed_comparisons
> >
> > Now I know Luabind can do more than OOLua and I would like to have
> results
> > which are fair, so if there is anything which you think I should be doing
> in
> > the tests or more defines to speed it up please do inform me.
>
> Which compiler are you using for this? I just tried you benchmark, after
> some fiddling to make it build, and while my results match yours for
> 0.8.1, for 0.9 they are completely different:
>
> Fiddling? I have built this using gcc on many flavours of linux, yet I
would be very grateful if you could inform me of any changes you had to
make.


> > OOLua access (average elapsed time):  0.698
> > OOLua passing derived to a function that wants a base  (average elapsed
> time):        0.719
> > OOLua passing derived with multiple bases to a function that wants a base
>  (average elapsed time):    0.747
> > OOLua virtual function  (average elapsed time):       0.338
> > OOLua pure virtual function  (average elapsed time):  0.341
> > OOLua - Lua function which takes a base and calls a virtual function
> (average elapsed time): 0.309
> > Swig access (average elapsed time):   1.582
> > Swig passing derived to a function that wants a base  (average elapsed
> time): 0.797
> > Swig passing derived with multiple bases to a function that wants a base
>  (average elapsed time):     0.835
> > Swig virtual function  (average elapsed time):        0.747
> > Swig pure virtual function  (average elapsed time):   0.755
> > Swig - Lua function which takes a base and calls a virtual function
> (average elapsed time): 0.74
> > Luabind access (average elapsed time):        1.089
> > Luabind passing derived to a function that wants a base  (average elapsed
> time):      0.659
> > Luabind passing derived with multiple bases to a function that wants a
> base (average elapsed time):   0.673
> > Luabind virtual function  (average elapsed time):     0.515
> > Luabind pure virtual function  (average elapsed time):        0.517
> > Luabind - Lua function which takes a base and calls a virtual function
> (average elapsed time): 0.502
>
> This is with:
>
>  gcc version 4.3.3 (Ubuntu 4.3.3-5ubuntu4)
>
> --
>
Yes that is vastly different, I will the check the new rc which a couple of
OS's I have here. The results I posted where from Visual Studio 9 sp1.

Liam

Re: [luabind] [ANN] luabind 0.9-rc1

From: Daniel Wallin <daniel@bo...> - 2009-12-10 21:27

liam mail wrote:
> Is there a packaging mistake for the 0.9rc1 zip file, I still see the
> directory inside named luabind-0.8.1 and the speed it almost the same?

Uh, no. Are you sure you didn't just download the wrong file? I just
double checked.

-- 
Daniel Wallin
BoostPro Computing
http://www.boostpro.com

Re: [luabind] [ANN] luabind 0.9-rc1

From: Daniel Wallin <daniel@bo...> - 2009-12-10 21:27

Tony Kostanjsek wrote:
>> So I tried the tar.gz packaged download and "simply dropping the  
>> source files in a Visual Studio project" yet I get errors.  
>> find_best_match.hpp seems to have been removed form luabind/detail
>> find_best_match.cpp
>> .\src\find_best_match.cpp(31) : error C2039: 'find_best_match' : is  
>> not a member of 'luabind::detail'
>> .\src\find_best_match.cpp(33) : error C2061: syntax error :  
>> identifier 'overload_rep_base'
>> .\src\find_best_match.cpp(44) : error C2065: 'num_overloads' :  
>> undeclared identifier
> 
> 
> I encountered the same error, but could build and run my project after  
> removing find_best_match.cpp. That was 0.9 from github, 1 or 2 weeks  
> ago, haven't tried the newest version yet. 

Yes, it's an old left over file. Will remove for release. Thanks.

-- 
Daniel Wallin
BoostPro Computing
http://www.boostpro.com

Re: [luabind] [ANN] luabind 0.9-rc1

From: liam mail <liam.list@go...> - 2009-12-10 21:45

Attachments: Message as HTML

2009/12/10 Daniel Wallin <daniel@...>

> Tony Kostanjsek wrote:
> >> So I tried the tar.gz packaged download and "simply dropping the
> >> source files in a Visual Studio project" yet I get errors.
> >> find_best_match.hpp seems to have been removed form luabind/detail
> >> find_best_match.cpp
> >> .\src\find_best_match.cpp(31) : error C2039: 'find_best_match' : is
> >> not a member of 'luabind::detail'
> >> .\src\find_best_match.cpp(33) : error C2061: syntax error :
> >> identifier 'overload_rep_base'
> >> .\src\find_best_match.cpp(44) : error C2065: 'num_overloads' :
> >> undeclared identifier
> >
> >
> > I encountered the same error, but could build and run my project after
> > removing find_best_match.cpp. That was 0.9 from github, 1 or 2 weeks
> > ago, haven't tried the newest version yet.
>
> Yes, it's an old left over file. Will remove for release. Thanks.
>
> --
> Daniel Wallin
> BoostPro Computing
> http://www.boostpro.com
>
>
> ------------------------------------------------------------------------------
> Return on Information:
> Google Enterprise Search pays you back
> Get the facts.
> http://p.sf.net/sfu/google-dev2dev
> _______________________________________________
> luabind-user mailing list
> luabind-user@...
> https://lists.sourceforge.net/lists/listinfo/luabind-user
>
Ok I ran the same test using g++ (Ubuntu 4.4.1-4ubuntu8) 4.4.1
I got the same results as you did, it seems you have made huge improvements
for the gcc version by for some reason mscv is lagging.

OOLua access (average elapsed time):    0.762
OOLua passing derived to a function that wants a base  (average elapsed
time):    0.879
OOLua passing derived with multiple bases to a function that wants a base
(average elapsed time):    0.856
OOLua virtual function  (average elapsed time):    0.346
OOLua pure virtual function  (average elapsed time):    0.356
OOLua - Lua function which takes a base and calls a virtual function
(average elapsed time): 0.379
Swig access (average elapsed time):    1.557
Swig passing derived to a function that wants a base  (average elapsed
time):    0.789
Swig passing derived with multiple bases to a function that wants a base
(average elapsed time):    0.835
Swig virtual function  (average elapsed time):    0.716
Swig pure virtual function  (average elapsed time):    0.724
Swig - Lua function which takes a base and calls a virtual function (average
elapsed time): 0.713
Luabind access (average elapsed time):    1.05
Luabind passing derived to a function that wants a base  (average elapsed
time):0.701
Luabind passing derived with multiple bases to a function that wants a base
(average elapsed time):    0.696
Luabind virtual function  (average elapsed time):    0.524
Luabind pure virtual function  (average elapsed time):    0.524
Luabind - Lua function which takes a base and calls a virtual function
(average elapsed time):

Re: [luabind] [ANN] luabind 0.9-rc1

From: Daniel Wallin <daniel@bo...> - 2009-12-10 21:50

liam mail wrote:
> Ok I ran the same test using g++ (Ubuntu 4.4.1-4ubuntu8) 4.4.1
> I got the same results as you did, it seems you have made huge improvements
> for the gcc version by for some reason mscv is lagging.

Here are my results for VC9.0:

> OOLua access (average elapsed time):    0.4719
> OOLua passing derived to a function that wants a base  (average elapsed time): 0.4188
> OOLua passing derived with multiple bases to a function that wants a base  (average elapsed time):      0.4328
> OOLua virtual function  (average elapsed time): 0.2046
> OOLua pure virtual function  (average elapsed time):    0.2156
> OOLua - Lua function which takes a base and calls a virtual function (average elapsed time): 0.2031
> Swig access (average elapsed time):     0.9297
> Swig passing derived to a function that wants a base  (average elapsed time): 0.4781
> Swig passing derived with multiple bases to a function that wants a base  (average elapsed time):       0.4922
> Swig virtual function  (average elapsed time):  0.4313
> Swig pure virtual function  (average elapsed time):     0.4344
> Swig - Lua function which takes a base and calls a virtual function (average elapsed time): 0.4297
> Luabind access (average elapsed time):  0.7953
> Luabind passing derived to a function that wants a base  (average elapsed time):         0.5734
> Luabind passing derived with multiple bases to a function that wants a base (average elapsed time):     0.5921
> Luabind virtual function  (average elapsed time):       0.4656
> Luabind pure virtual function  (average elapsed time):  0.4719
> Luabind - Lua function which takes a base and calls a virtual function (average elapsed time): 0.4093

Are you sure you built the benchmark correctly with 0.9?

-- 
Daniel Wallin
BoostPro Computing
http://www.boostpro.com

Re: [luabind] [ANN] luabind 0.9-rc1

From: Dean Harding <dean-ml@co...> - 2009-12-10 22:10

> Are you sure you built the benchmark correctly with 0.9?

Perhaps it was compiled without specifying _SECURE_SCL=0? Just a guess...

Dean.

Re: [luabind] [ANN] luabind 0.9-rc1

From: liam mail <liam.list@go...> - 2009-12-10 22:39

Attachments: Message as HTML

>
> Are you sure you built the benchmark correctly with 0.9?
>
> I must of messed the build up or had something running in the background
that I was not aware of.

OOLua access (average elapsed time):    0.7937
OOLua passing derived to a function that wants a base  (average elapsed
time):0.836
OOLua passing derived with multiple bases to a function that wants a base
(average elapsed time):      0.8172
OOLua virtual function  (average elapsed time): 0.3219
OOLua pure virtual function  (average elapsed time):    0.3281
OOLua - Lua function which takes a base and calls a virtual function
(average elapsed time): 0.3125
Swig access (average elapsed time):     1.4828
Swig passing derived to a function that wants a base  (average elapsed
time):0.7703
Swig passing derived with multiple bases to a function that wants a base
(average elapsed time):       0.7829
Swig virtual function  (average elapsed time):  0.7078
Swig pure virtual function  (average elapsed time):     0.7063
Swig - Lua function which takes a base and calls a virtual function (average
elapsed time): 0.6984
Luabind access (average elapsed time):  1.1156
Luabind passing derived to a function that wants a base  (average elapsed
time): 0.8375
Luabind passing derived with multiple bases to a function that wants a base
(average elapsed time):     0.8625
Luabind virtual function  (average elapsed time):       0.5859
Luabind pure virtual function  (average elapsed time):  0.5828
Luabind - Lua function which takes a base and calls a virtual function
(average elapsed time): 0.5265

Re: [luabind] [ANN] luabind 0.9-rc1

From: Colm Sloan <colmsloan@gm...> - 2009-12-11 00:48

Attachments: Message as HTML

Fantastic work. I didn't realize how much had been done. Where can I read
more about these two points?:

 * Automatically perform down-casts as well as up-casts when converting
  arguments from Lua to C++.

 * When converting from C++ to Lua, polymorphic instances are
  automatically cast to their most derived type.

I'd like to give these a try.

Re: [luabind] [ANN] luabind 0.9-rc1

From: Daniel Wallin <daniel@bo...> - 2009-12-13 01:03

Colm Sloan wrote:
> Fantastic work. I didn't realize how much had been done. Where can I read
> more about these two points?:

Both of these features are actually not documented due to lack of time.
Here's a short explanation though:

>  * Automatically perform down-casts as well as up-casts when converting
>   arguments from Lua to C++.

Rather than just cast up to base types, we search the graph in both
directions. See example below.

>  * When converting from C++ to Lua, polymorphic instances are
>   automatically cast to their most derived type.

This means that code like this:

  struct X
  {
      virtual ~X() {}
  };

  struct Y : X
  {
      void f();
  };

  X* make()
  {
      return new Y;
  }

  void take(Y*);

  ...

    class_<X>("X"),
    class_<Y, X>("Y")
      .def("f", &Y::f, adopt(result)),
    def("make", &make),
    def("take", &take)

.. and later in Lua:

  x = make()
  x:f()

will work. The dynamic type of the instance returned from make() will be
used to determine the class.

  take(x)

will also work, because luabind can cast the held X* down to a Y*.

HTH,
-- 
Daniel Wallin
BoostPro Computing
http://www.boostpro.com

Re: [luabind] [ANN] luabind 0.9-rc1

From: <fuscated@gm...> - 2010-01-02 21:00

Attachments: luabind_0.9rc1_build.patch

On Thu, 10 Dec 2009 13:19:11 +0100
Daniel Wallin <daniel@...> wrote:

> I'm happy to announce the first RC of luabind 0.9. There have been a lot
> of improvements since 0.8.1. Among the more interesting ones are the
> automatic handling of down casts, as well as better smart pointer
> support, and of course the improved performance.
> 
> Many thanks to Athena Capital Research who funded a large part of the
> development for this release.
> 
> Please test and report problems!

Hello,
And thank you for your hard work.

I've updated luabind to 0.9rc1 and found that the LUABIND_NO_EXCEPTIONS
was not respected everywhere in the code.
I've attached a patch that make the thing compile (also it seems to work).
Is it possible to make luabind work without exceptions again?

Best regards,
Teodor Petrov

Re: [luabind] [ANN] luabind 0.9-rc1

From: Daniel Wallin <daniel@bo...> - 2010-01-05 09:09

fuscated@... wrote:
> On Thu, 10 Dec 2009 13:19:11 +0100
> Daniel Wallin <daniel@...> wrote:
> 
>> I'm happy to announce the first RC of luabind 0.9. There have been a lot
>> of improvements since 0.8.1. Among the more interesting ones are the
>> automatic handling of down casts, as well as better smart pointer
>> support, and of course the improved performance.
>>
>> Many thanks to Athena Capital Research who funded a large part of the
>> development for this release.
>>
>> Please test and report problems!
> 
> Hello,
> And thank you for your hard work.
> 
> I've updated luabind to 0.9rc1 and found that the LUABIND_NO_EXCEPTIONS
> was not respected everywhere in the code.
> I've attached a patch that make the thing compile (also it seems to work).
> Is it possible to make luabind work without exceptions again?

Thanks. I've applied your patch.

-- 
Daniel Wallin
BoostPro Computing
http://www.boostpro.com