www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - inheritance syntax

reply Marc <Marc_member pathlink.com> writes:
Hello,

I'm wondering why D doesn't use the Java-style syntax for class inheritance
(extends, implements), because I thought D's purpose was to have the clearest
possible syntax.
In its current state, it seems to me it's not possible to determine visually the
type of inheritance, eg.:

class Y: A, B, C, D
{
}

is it immediatly possible to say which of A, B, C or D are interfaces or classes
?

bye,
Marc
May 18 2004
next sibling parent reply J Anderson <REMOVEanderson badmama.com.au> writes:
Marc wrote:

Hello,

I'm wondering why D doesn't use the Java-style syntax for class inheritance
(extends, implements), because I thought D's purpose was to have the clearest
possible syntax.
In its current state, it seems to me it's not possible to determine visually the
type of inheritance, eg.:

class Y: A, B, C, D
{
}

is it immediatly possible to say which of A, B, C or D are interfaces or classes
?

bye,
Marc

  
D copies a lot of syntax from C++ which makes it easy for people who use the worlds most popular language to learn D. -- -Anderson: http://badmama.com.au/~anderson/
May 18 2004
parent reply marc.noirotREM OVEfleximage.fr writes:
IMHO, the reason you're giving me is not satisfying.
D is supposed to be a language that solves the flaws of C++.
That was successfully done with templates, and you won't say that the new syntax
is familiar to C++ developers ;)
I think the ':' inheritance syntax is one of C++ flaws. I'm a C++ supporter, but
I find that the words 'extends' and 'implements' are really expressive and
meaningful (and leave no place for contextual interpretation).
So, if i follow your argument, why not having D support both syntaxes, so C++
and Java developers will both feel at ease with it ?

In article <c8clja$2dn$1 digitaldaemon.com>, J Anderson says...
Marc wrote:

Hello,

I'm wondering why D doesn't use the Java-style syntax for class inheritance
(extends, implements), because I thought D's purpose was to have the clearest
possible syntax.
In its current state, it seems to me it's not possible to determine visually the
type of inheritance, eg.:

class Y: A, B, C, D
{
}

is it immediatly possible to say which of A, B, C or D are interfaces or classes
?

bye,
Marc

  
D copies a lot of syntax from C++ which makes it easy for people who use the worlds most popular language to learn D. -- -Anderson: http://badmama.com.au/~anderson/
May 18 2004
next sibling parent reply J Anderson <REMOVEanderson badmama.com.au> writes:
marc.noirotREM OVEfleximage.fr wrote:

 So, if i follow your argument, why not having D support both syntaxes, 
 so C++
 and Java developers will both feel at ease with it ?
  
Obviously you can't have to many alias for things in languages, it makes the language to broad (language design 1.0 -> unit). You have to pick one direction and stick with it. Why not also make blocks { } begin and end? -- -Anderson: http://badmama.com.au/~anderson/
May 18 2004
parent reply marc.noirot REMOVEfleximage.fr writes:
Obviously you can't have to many alias for things in languages, it makes 
the language to broad (language design 1.0 -> unit).  You have to pick 
one direction and stick with it.  Why not also make blocks { } begin and 
end?
Of course, i agree with you, i was actually being half-sarcastic. But well, let's elude that point... I think ':' is acceptable in C++ because it doesn't have the builtin notion of interface. But like in Java, D's class inheritance and interface implementation are two different notions, so why writing them as if they were the same thing ? If we make abstraction of the fact that D wants to stick with C++ syntax, what is the best for someone trying to learn a language from scratch ? A perlish symbol like ':' expressing 2 notions ? Or two words like 'extends' and 'implements' that leave no doubt about the context ? Thanks for your patience in answering my irritating questions :P Marc
May 18 2004
parent reply J Anderson <REMOVEanderson badmama.com.au> writes:
marc.noirot REMOVEfleximage.fr wrote:

Obviously you can't have to many alias for things in languages, it makes 
the language to broad (language design 1.0 -> unit).  You have to pick 
one direction and stick with it.  Why not also make blocks { } begin and 
end?

    
Of course, i agree with you, i was actually being half-sarcastic. But well, let's elude that point... I think ':' is acceptable in C++ because it doesn't have the builtin notion of interface. But like in Java, D's class inheritance and interface implementation are two different notions, so why writing them as if they were the same thing ?
Often you only want to implement one interface/class and users don't care if its a class or interface. I believe it helps to keep things like this generic. You don't need to search the docs to find out if something is an interface or class. If you need to use more then one interface, then you can search the docs, if nessary. Furthermore if the developer wishes to change an interface to a class (or visa versa), they won't have maintance problems in D. If you really want this documentation you can put a comment in.
If we make abstraction of the fact that D wants to stick with C++ syntax, what
is the best for someone trying to learn a language from scratch ?
A perlish symbol like ':' expressing 2 notions ? Or two words like 'extends' and
'implements' that leave no doubt about the context ?

Thanks for your patience in answering my irritating questions :P
Marc
  
No worries. -- -Anderson: http://badmama.com.au/~anderson/
May 18 2004
next sibling parent reply Marc <marc.noirotMOVE REfleximage.fr> writes:
ok, what about this ? :

//interclace.d
import std.c.stdio;

//*
version=1;
//*/

version (1) {
interface A {
char [] toString();
}
}
else {
class A {
char [] toString() {
return "3";
}
}
}

/* if version = 1, super.toString == Object.toString,
else super.toString == A.toString */
class B: A {
char [] toString() {
return super.toString() ~ "1";
}
}

void main() {
B b = new B;
printf("B: %s\n", cast(char *)b.toString()); // what does it print ?
}

//EOF

the output of b.toString() is not 'really' predictable if version = 1
(Object.toString seems to print "Object", but is it implementation dependant
?)although both versions are valid and compile fine.
weirdness, weirdnesss, doesn't it look like a potential source of bugs that
could be avoided with explicit "implements" or "extends" ?
May 18 2004
parent J Anderson <REMOVEanderson badmama.com.au> writes:
Marc wrote:

ok, what about this ? :

//interclace.d
import std.c.stdio;

//*
version=1;
//*/

version (1) {
interface A {
char [] toString();
}
}
else {
class A {
char [] toString() {
return "3";
}
}
}

/* if version = 1, super.toString == Object.toString,
else super.toString == A.toString */
class B: A {
char [] toString() {
return super.toString() ~ "1";
}
}

void main() {
B b = new B;
printf("B: %s\n", cast(char *)b.toString()); // what does it print ?
}

//EOF

the output of b.toString() is not 'really' predictable if version = 1
(Object.toString seems to print "Object", but is it implementation dependant
?)although both versions are valid and compile fine.
weirdness, weirdnesss, doesn't it look like a potential source of bugs that
could be avoided with explicit "implements" or "extends" ?


  
This is one reason I like D's form of interfaces. Later in a maintanace cycle it may be decided that interface A is better off as an abstract class (or whatever). If you had implements and extends you'd need too go to every one of your clients code and modify it to work. Besides you can easily produce the error message you want by adding a stub constructor and calling super. -- -Anderson: http://badmama.com.au/~anderson/
May 18 2004
prev sibling parent reply "Matthew" <matthew.hat stlsoft.dot.org> writes:
"J Anderson" <REMOVEanderson badmama.com.au> wrote in message
news:c8d6go$vpm$1 digitaldaemon.com...
 marc.noirot REMOVEfleximage.fr wrote:

Obviously you can't have to many alias for things in languages, it makes
the language to broad (language design 1.0 -> unit).  You have to pick
one direction and stick with it.  Why not also make blocks { } begin and
end?
Of course, i agree with you, i was actually being half-sarcastic. But well, let's elude that point... I think ':' is acceptable in C++ because it doesn't have the builtin notion of interface. But like in Java, D's class inheritance and interface implementation are two different notions, so why writing them as if they were the same thing ?
Often you only want to implement one interface/class and users don't care if its a class or interface. I believe it helps to keep things like this generic. You don't need to search the docs to find out if something is an interface or class. If you need to use more then one interface, then you can search the docs, if nessary. Furthermore if the developer wishes to change an interface to a class (or visa versa), they won't have maintance problems in D. If you really want this documentation you can put a comment in.
Excellent point. And I should have thought about this before answering the other post in this thread. DTL relies on some types deriving from template parameters. It would be a restriction if they were required to stipulate whether this parameterising parent was either class or interface. So I say No! to the idea.
Jun 04 2004
parent Marc <marc.noirotNO flexSPAMimage.fr> writes:
 Furthermore if the
 developer wishes to change an interface to a class (or visa versa), they
 won't have maintance problems in D.
that's not always true... if an interface becomes a class, then a class extending some superclass AND implementing that former interface will need to be redesigned, because of multiple inheritence not being supported. if a class becomes an interface, there are big chances that not all methods are implemented by a class using it, so the developer will have to provide default implementations for methods he doesn't care about. In a pure OOP approach, I can't admit that interfaces and classes are considered the same. Here are some URLs where the topic is discussed : http://www.javaworld.com/javaworld/javaqa/2001-04/03-qa-0420-abstract.html http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vbcon/html/vbconAbstractClassesVersusInterfaces.asp http://c2.com/cgi/wiki?DontDistinguishBetweenClassesAndInterfaces http://c2.com/cgi/wiki?InterfacesVsClasses regards, Marc
Jun 08 2004
prev sibling next sibling parent Andy Friesen <andy ikagames.com> writes:
marc.noirotREM OVEfleximage.fr wrote:
 IMHO, the reason you're giving me is not satisfying.
 D is supposed to be a language that solves the flaws of C++.
 That was successfully done with templates, and you won't say that the new
syntax
 is familiar to C++ developers ;)
 I think the ':' inheritance syntax is one of C++ flaws. I'm a C++ supporter,
but
 I find that the words 'extends' and 'implements' are really expressive and
 meaningful (and leave no place for contextual interpretation).
 So, if i follow your argument, why not having D support both syntaxes, so C++
 and Java developers will both feel at ease with it ?
It is not a flaw at all, merely a different syntax, as it does not effect *what* can be expressed in D, merely how it is expressed. -- andy
May 18 2004
prev sibling parent "Matthew" <matthew.hat stlsoft.dot.org> writes:
I think there's mileage in this idea, but I suspect it's too late to get
anything
done about. One might also say it's not really important enough to warrant the
changes at this point.

<marc.noirotREM OVEfleximage.fr> wrote in message
news:c8coe5$6pj$1 digitaldaemon.com...
 IMHO, the reason you're giving me is not satisfying.
 D is supposed to be a language that solves the flaws of C++.
 That was successfully done with templates, and you won't say that the new
syntax
 is familiar to C++ developers ;)
 I think the ':' inheritance syntax is one of C++ flaws. I'm a C++ supporter,
but
 I find that the words 'extends' and 'implements' are really expressive and
 meaningful (and leave no place for contextual interpretation).
 So, if i follow your argument, why not having D support both syntaxes, so C++
 and Java developers will both feel at ease with it ?

 In article <c8clja$2dn$1 digitaldaemon.com>, J Anderson says...
Marc wrote:

Hello,

I'm wondering why D doesn't use the Java-style syntax for class inheritance
(extends, implements), because I thought D's purpose was to have the clearest
possible syntax.
In its current state, it seems to me it's not possible to determine visually
the
type of inheritance, eg.:

class Y: A, B, C, D
{
}

is it immediatly possible to say which of A, B, C or D are interfaces or
classes
?

bye,
Marc
D copies a lot of syntax from C++ which makes it easy for people who use the worlds most popular language to learn D. -- -Anderson: http://badmama.com.au/~anderson/
Jun 04 2004
prev sibling parent reply Genki Takiuchi <takiuchi ntechnology.co.jp> writes:
Marc wrote:
 Hello,
 
 I'm wondering why D doesn't use the Java-style syntax for class inheritance
 (extends, implements), because I thought D's purpose was to have the clearest
 possible syntax.
 In its current state, it seems to me it's not possible to determine visually
the
 type of inheritance, eg.:
 
 class Y: A, B, C, D
 {
 }
 
 is it immediatly possible to say which of A, B, C or D are interfaces or
classes
 ?
 
 bye,
 Marc
 
 
Hi. How about writing like this? class Foo : SuperClass , InterfaceOne , InterfaceTwo , InterfaceThree { }
May 18 2004
parent "Matthew" <matthew.hat stlsoft.dot.org> writes:
"Genki Takiuchi" <takiuchi ntechnology.co.jp> wrote in message
news:c8ebp7$2uh8$1 digitaldaemon.com...
 Marc wrote:
 Hello,

 I'm wondering why D doesn't use the Java-style syntax for class inheritance
 (extends, implements), because I thought D's purpose was to have the clearest
 possible syntax.
 In its current state, it seems to me it's not possible to determine visually
the
 type of inheritance, eg.:

 class Y: A, B, C, D
 {
 }

 is it immediatly possible to say which of A, B, C or D are interfaces or
classes
 ?

 bye,
 Marc
Hi. How about writing like this? class Foo : SuperClass , InterfaceOne , InterfaceTwo , InterfaceThree { }
I always write inheritance relationships as class Foo : SuperClass , InterfaceOne , InterfaceTwo , InterfaceThree { ... } However, I think you may have something with your proposal. We should adopt the convention that the parent class always comes first, and if that's on the same line, then anything below that is an interface. However, this fails when we want the parent(s) to be dependent on template parameters. Nonetheless, for concrete types, it'd be nice to try and at least follow the principal that parent classes come first.
Jun 04 2004