Memory management in C++ Extension
-
Hello,
I've been playing with C/C++ extension starting from the Simple Extension example and I'm currently facing some challenges dealing with memory.
My goal is to build a tree in memory matching like the Outliner, and load the geometry to do some analysis.
I've built a class with a tree structure named "Coll" of pointer type "Entity", who has a std::vector of pointer of "Points". I always got a memory corruption error whenever I want to reload and delete some allocated memory.
What am I doing wrong ?Olivier
-
Hard to tell without seeing any code. If it proprietary code or something you can add to GitHub?
What are you point to anyway?
Are you wrapping your C/C++ object in the Ruby objects?
Do you make sure to account for Ruby exceptions that will long jump and possibly make you skip some cleanup methods?Are you using old-school pointers or smart pointers?
-
It is really hard to tell without more information of the code.
-
I'm trying to see if I can get some collision detection. I've only wrap one object to get access to the active_model. Visual Studio allows to go in debug mode, so I don't see any exception happening. I tried smart pointers but got back to regular pointers.
-
I haven't had time to try this, but just some first glance generic observations:
// Coll.h typedef unsigned long VALUE; typedef unsigned long ID;
Why are you typedefing these? It'd be much safer to include the Ruby headers for this - as the type for VALUE and ID might vary from Ruby version to Ruby version.
void Coll;;Init_SkpCol() { VALUE rModule = rb_define_module("CColl"); rCollClass = rb_define_class_under(rModule, "Coll", rb_cObject); rb_define_method(rCollClass, "initialize", VALUEFUNC(_wrap_initialize), 1); rb_define_singleton_method(rCollClass, "new", VALUEFUNC(_wrap_new), -1); rb_define_method(rCollClass, "load_model", VALUEFUNC(_wrap_load_model), 1); }
Using
new
to allocate C data is a very old way of doing it and causes some complications when one want to create new instances. SketchUp has been using this way of initializing as well, which caused me some headache a while ago.
You can read it on StackOverflow along with links that describes the new way of doing this: http://stackoverflow.com/a/9946578/486990This doesn't address your original questions, but I just wanted to let you know.
Can you describe more where you where seeing your errors etc? It's a bit daunting to jump straight into code you know nothing about.
-
I first typedefed VALUE and ID because of some include issues with ruby.h there. But it seems to be over now.
I changed the code to allocate the wrapped C++ class using
rb_define_alloc_func
but that didn't solved the issue.I get the memory error whenever I call load_model a second time, and need to do before some cleaning in the Reset() method of dynamically allocated memory (EntityInfo) in the tree.
-
Should it be "<" instead of "<=" ?
for(int i = 0; i **%(#0000FF)[<]** EntityTree.depth(it2)-rootdepth; ++i)
The original code:
void Coll;;Reset(void) { TreeEntity;;post_order_iterator it2 = EntityTree.begin_post(); TreeEntity;;post_order_iterator end2 = EntityTree.end_post(); int rootdepth = EntityTree.depth(it2); while(it2 != end2) { for(int i = 0; i <= EntityTree.depth(it2)-rootdepth; ++i) if (it2.node->data) delete it2.node->data; ++it2; } }
-
@olilej said:
Visual Studio allows to go in debug mode, so I don't see any exception happening.
You don't see the crash when you attach Visual Studio and run it?
-
I finally found my silly mistake using static keyword here, it was getting in the way of the GC.
%(#404080)[VALUE CRubyUtils::GetDAttribute(VALUE obj, VALUE att) { ~~static~~ VALUE getatt = rb_intern("get_attribute"); ~~static~~ VALUE dicatt = rb_str_new2("dynamic_attributes"); return rb_funcall(obj, getatt, 2, dicatt, att); }]
Thanks for your support, might see you at Basecamp.
-
Hmm.. did you have to remove static for the rb_intern call as well? I can understand that making the reference to the string object you created static could cause problems, but the rb_intern call?
-
From what I understand, static function variables persists out of the scope of the function and are only allocated once in a specific memory area.
As this memory was allocated using a VALUE pointer unknown from Ruby, the Garbage Collector cleaned it up and I ended up with a Memory Access Violation.So I would say, it is not a good idea to use a static variable of type VALUE, unless you find a way to tell Ruby GC by marking it.
-
No, thats not quite right.
You've made those variables static (FWIW, for no benefit) - but because their scope is limited to that function, nothing outside can access them so you should be fine with that.
The golden rule is never hang onto any VALUE reference beyond the scope of your functions.
When you get more experienced with Ruby extensions, you may want to look into using rb_gc_register_address() but I would strongly suggest you stick to never holding VALUE references unless you really understand whats going on under the hood.
Adam
-
No, true values (as opposed to references) - IDs etc - are fine to cache.
So method selectors can happily be cached once at startup. Just as Thomthom says, avoiding the additional function calls is a "Good Thing (tm)".
-
Thanks for clarifying Adam.
I have seen ID and VALUE mixed in Ruby C Extensions - which there is no compiler warning about since they are typedefs of the same type:
http://rxr.whitequark.org/mri/source/include/ruby/ruby.h#082
-
rb_intern doesn't return VALUE, it returns ID. Adam, would the same issues for VALUE be valid for ID?
I've used to make static const out of ID's I use a lot for easy access and avoid Ruby lookup up for every function call.
Advertisement