Talk:Trait

From APIDesign

Revision as of 08:34, 18 September 2012 by JaroslavTulach (Talk | contribs)
Jump to: navigation, search

Comments on Trait <comments />

Contents

Miles Elam said ...

Yes, this is indeed possible in C++ and is, in fact, used extensively in the C++ standard library (aka STL). For a prime example, look no further than std::string or the various pluggable memory allocators. The example given in the article appears to these eyes as one of a C programmer trying to make C++ do things like C and failing. To be more precise, if one is accessing people objects by iterator, why would a raw pointer to a person need to be manipulated in this way? In addition, what happens if the object must be accessed in multiple ways, e.g., exists in both a normal list and a sorted list (or multiple sorted lists). The C method falls down as there is no single pair of *next and *prev but rather multiple.

Don't get me wrong, C definitely has its uses. Its relative simplicity for one. However, C++'s generic algorithms and data structures should not be discarded so lightly.

--Miles Elam 20:26, 4 September 2012 (CEST)

I am primarily interested in properly typing the multiple class encapsulation case. E.g. having prev/next field in the item class and manipulating them in only by the list. Moreover I'd like to write this (and type this) in a generic way. Looks like it will be possible to do it in C++, but the solution will definitely not be like in Java/Scala - rather upside-down...

--JaroslavTulach 22:25, 5 September 2012 (UTC)

jtulach said ...

The article is more about STL overhead. I personally don't like and don't use STL. I think following C++ implementation is as fast as implementation in C. 

class ll_item;
 
class llist {
  public:
    void add(ll_item& p);
    void remove(ll_item& p);
 
    llist(void);
   ~llist();
};
 
class ll_item {
    friend llist;
  private:
    ll_item* next;
    ll_item* prev;
};
 
 
class person_item : public ll_item {
  protected:
    int age;
    const char* name;
 
  public:
    person_item(int age, const char* name);
};
 
class animal_item : public ll_item {
  protected:
    const char* name;
 
  public:
    animal_item(const char* name);
};
 
int main(int argc, char* argv[])
{
  person_item a(10, "Ben");
  person_item b(20, "Nora");
  person_item c(30, "John");
 
  animal_item x("Fifi");
  animal_item y("Bobika");
  animal_item z("Bill");
 
  llist l;
  l.add(b);
  l.add(c);
  l.add(a);
  l.add(x);
  l.add(y);
  l.add(z);
}

--jtulach 21:07, 11 September 2012 (CEST)

Right, now the question is how to generify this (as the example above does not feel type safe enough) - e.g. how to turn into a C++ template?

I don't want to have list of items, but rather list of persons and another list of animals. When I ask for an item from each list, I would expect I get a person in the former case and an animal in the latter. Right now I just get a ll_item. That is efficient, but not really typesafe.

I have an unfinished prototype with template and compared to Java or Scala it feels a bit upside down. Interesting clash of cultures.

--JaroslavTulach 08:26, 12 September 2012 (UTC)

jtulach said ...

I don't know why do you think this is not type safe? You cannot add anything which is not inherited from ll_item. If you have enabled RTTI, you can dynamic cast items from list. 

class persons_list : public llist {
  public:
    person_item *get(int idx)
    {
      return dynamic_cast<person_item *>(llist::get(idx));
    }
};

--jtulach 10:58, 13 September 2012 (CEST)

The question Why this is not type-safe? may reveal the progress OOP made during last twenty years. In the middle of nineties Java was claimed a safe language as it could not cause segmentation faults. The C++'s dynamic_cast is a safety of similar kind. It is a runtime check. It is not bad thing, but as I am coding in Java most of the time, I was not calling for something as basic.

When I mentioned type safety, I meant compile type safety. The above dynamic cast does not guarantee that I cannot add animal_item into the llist and try to obtain person_item. To achieve such kind of safety one needs a form of algebraic types (nicely illustrated in the revolutionary essay). Both Scala and Java (since version 1.5) provide a way to marry algebraic types with OOP inheritance and thus provide compile time safety (as code snippets for the direct linked list show).

--JaroslavTulach 07:42, 17 September 2012 (UTC)

jtulach said ... 

template<class ITEM>
class llist {
  protected:
        ITEM *first;
  public:
	void add(ITEM& p)
        {
          p.prev=NULL;
          p.next=first;
          first=&p;
        }
	void remove(ITEM& p)
        {
          ITEM *pr=p.prev;
          ITEM *nx=p.next;
          if(pr) pr->next=nx;
          else first=nx;
          if(nx) nx->prev=pr;
        }
 
        ITEM *get(int idx)
        {
          ITEM *i=first;
          while((i)&&(idx-->0))
            i=i->next;
 
          return i;
        }
 
        llist(void) { first=NULL; };
        ~llist() { };
};
 
// not inherited, requires to define next and prev properties
class person_item  {
  protected:
  int age;
  const char* name;
 
  public:
  person_item* next;
  person_item* prev;
 
  person_item (int age, const char* name) { this->age=age; this->name=name; }
};
 
 
template<class ITEM>
class ll_item  {
  friend llist <ITEM>;
private:
  ITEM* next;
  ITEM* prev;
};
 
// inherited from ll_item template which defines next and prev for me
class animal_item : public ll_item <animal_item>{
  protected:
  const char* name;
 
  public:
 
  animal_item(const char* name) : ll_item <animal_item>() { this->name=name; }
};
 
int main(int argc, char* argv[])
{
  person_item a(10, "Ben");
  person_item b(20, "Nora");
  person_item c(30, "John");
 
  animal_item x("Fifi");
  animal_item y("Bobika");
  animal_item z("Bill");
 
  llist <person_item> persons;
  persons.add(b);
  persons.add(c);
  persons.add(a);
//  persons.add(x); - compilation fails
 
  llist <animal_item> animals;
  animals.add(x);
  animals.add(y);
  animals.add(z);
}

--jtulach 14:04, 17 September 2012 (CEST)

Right. This is it.

I didn't know about the option used when defining person_item - e.g. to just expose the next and prev publicly. Not that I'd prefer that solution (as it breaks the encapsulation), but shows the power of C++ always recompiled templates. Java's generics are compiled only once and need to use the class animal_item : public ll_item <animal_item> approach.

Thanks. I'll publish the code somehow, sometimes on the blogs.

--JaroslavTulach 08:34, 18 September 2012 (UTC)

Retrieved from "http://wiki.apidesign.org/wiki/Talk:Trait"
Personal tools
buy