Difference between revisions of "Programming in D for Delphi Programmers"

From D Wiki
Jump to: navigation, search
(The const-ness gate)
m (The const-ness gate: - transitive-const)
Line 702: Line 702:
 
<code>Error: cannot modify const expression aFoo.member</code>
 
<code>Error: cannot modify const expression aFoo.member</code>
  
because you try to modify a const parameter. D2 is more strict in regard of the '''const''' parameter storage class.
+
because you try to modify a const parameter. D2 is more strict in regard of the '''const''' parameter storage class. For a class instance or a struct the [http://dlang.org/const-faq.html#transitive-const constness is transitive].
  
 
You'll encounter the same '''strict''' behavior with [http://dlang.org/function.html#function-safety modular attributes] (@safe, @trusted, @system, @nogc).
 
You'll encounter the same '''strict''' behavior with [http://dlang.org/function.html#function-safety modular attributes] (@safe, @trusted, @system, @nogc).

Revision as of 09:51, 3 March 2015

Introduction

Delphi and Pascal come from another language family than D. D uses the C syntax. In short, a C syntax uses more symbols and more operators.

The following document develops a few very specific points over which D and Delphi/Pascal have some similarities or ambiguities.

Note that it's not an exhaustive comparison: for simple things such as loops, conditional statments, control flow, it's advisable to learn a bit the basis from the examples, the manual, Rosetta code, etc.

Types equivalence

Pascal D notes
Pointer void*
Byte ubyte 8 bit unsigned integer. note the ambiguity when coming from Pascal
ShortInt byte 8 bit signed integer
Word ushort 16 bit unsigned integer
SmallInt short 16 bit signed integer
DWord / Cardinal uint 32 bit unsigned integer
Integer int 32 bit signed integer
UInt64 ulong 64 bit unsigned integer
Int64 long 64 bit signed integer
Single float IEEE 754 32 bit floating point
Double double IEEE 754 64 bit floating point
NativeUInt size_t library type, either an alias to a unsigned 32 bit integer or to a unsigned 64 bit integer
NativeInt ptrdiff_t library type, either an alias to a signed 32 bit integer or to a signed 64 bit integer

The object model

The object model is equivalent. Multiple inheritence of classes is not allowed but can be achieved with interfaces.

  • Delphi, Pascal
type ISome = interface
    procedure blah();
end;

type TThing = class
end;

type  TSomething = class(TThing, ISome)
  procedure blah();
end;
  • The D2 way:
interface ISome(){
    void blah();
}

class TThing{
}

class TSomething: TThing, ISome{
    void blah(){}
}

However some slight difference exists in this model. The Pascal virtual method attribute does not exist in D. In Pascal, virtual methods must be marked, in D, every method is virtual unless it's marked as final.

for example, in Pascal:

type TFoo = class(TObject)
private
  procedure something1(); virtual;
  procedure something2();
end;

and the D equivalent:

class Foo : Object {
private:
  void something1();
  final void something2();
}

Memory management

Memory management is one of the biggest difference encountered when learning D with a Pascal background. This is less an aspect of the language than from the runtime. D uses a garbage collector (GC) while Pascal and traditionnal Delphi (before the switch to llvm ARC) uses manual memory management.

For example the following program in Pascal leaks a TObject.

program program1;
var
  obj: TObject;
begin
  obj := TObject.Create;
  // obj.Free; // leak
end;

While the D equivalent does not, because the Object reference is collected:

void main(string[] args){
  Object obj = new Object; // new causes collection
  // the GC will delete Obj automatically
}

However several technics exist to override the GC. For example it's possible to allocate a heap chunk which is not known by the GC, in the same fashion as done in Pascal with GetMem, ReallocMem and FreeMem:

import std.c.stdlib;
auto chunk = malloc(4096);
scope(exit) free(chunk);
// some code...

which is equivalent to

var
  chunk: Pointer;
begin
  chunk := GetMem(4096);
  try
    // some code...
  finally
    FreeMem(chunk);
  end;
end;

Allocating an Object instance or a struct in a heap zone which is not known by the GC is also possible. A few years back, it was possible to write a custom class allocator/deallocator (instrinsicly called with new and delete) but it's now deprecated. However there is a new way to do the same:

// GC-free struct: auto s = construct!S(a,b,c,...); // instead of auto s = new S(a,b,c...)
ST * construct(ST, A...)(A a)
if(is(ST==struct))
{
    import std.conv : emplace;
    auto size = ST.sizeof;
    auto memory = malloc(size)[0 .. size];
    if(!memory) throw new Exception("Out of memory");
    return emplace!(ST, A)(memory, a);
}

// GC-free class instance: auto c = construct!C(a,b,c,...); // instead of auto c = new C(a,b,c...)
CT construct(CT, A...)(A a) 
if (is(CT == class))
{
    import std.conv : emplace;
    auto size = __traits(classInstanceSize, CT);
    auto memory = malloc(size)[0 .. size];
    if(!memory) throw new Exception("Out of memory");
    return emplace!(CT, A)(memory, a);
}

// destroy a GC-free class: UFCS able, c.destruct;
static void destruct(T)(ref T instance) 
if (is(T == class) || (isPointer!T && is(PointerTarget!T == struct)))
{
    if (!instance) return;
    destroy(instance);
    instance = null;
}

You may have the idea to encapsulate the class allocator in a static class function, to emulate the Pascal syntax

myInstance = TSomething.create;

For example, using our previous function:

class Something
{
    static typeof(this) Create(A...)(A a){
        return construct!(typeof(this))(a);
    }
}

But this doesn't work because the static method is not redefined in the descendant. For example:

class SomethingElse: Something {}
SomethingElse se = SomethingElse.Create; // wont compile, Create() returns a Something not a SomethingElse

Templates

Traditional Delphi and Pascal didn’t supported templates. Delphi supports them since D2009, using the most common syntax: the left/right angle brackets (declaration and instantiation of a template).

D uses parens for the declaration and the exclamation mark (!) for the instantiation. Parens are used after the ! if the template expects several parameters.

  • Delphi or FPC with the dialect {$Mode Delphi}
type foo<T,G,A> = class
end;
  
type fooccc = class(foo<char, char, char>)
end;
    
var
  bar: foo<integer,single,double>;
  • FPC with the dialect {$Mode objfpc} (aka object Pascal)

Faithfully to the Pascal tradition, the objfpc dialect recquires two explicit keywords for the template declaration and the template instantiation: generic and specialize:

type generic foo<T,G,A> = class
end;

type fooccc = class(specialize foo<char, char, char>)
end;

var
  bar: specialize foo<integer,single,double>;
  • the D2 way:
class foo(T,G,A){
}
   
class fooccc: foo!(char, char, char){
}
   
foo!(int, float, double) bar;

To go further with the D2 templates, it's important to note that the syntax used in the previous example matches to what is called an eponymous template. An eponymous template is a template which contains only one member using the same identifier.

for example, the previous class foo is simplified form of:

template foo(T,G,A){
    class foo{}
}
// <template_name>!(params...).template_member
class fooccc: foo!(char, char, char).foo {}

This reveals that D templates are more powerfull than their Pascal equivalent because almost everything (alias, function, class, struct, interface, etc.) can be enclosed by a definition of parameters:

template tmp(T,C){

    void functA(){/*can use the parameters A and C*/}

    C functA(){/*can use the parameters A and C*/ return C.init;}

    T functA(){/*can use the parameters A and C*/ return T.init;}

    interface intfA{
        void methodA(T aT);
        void methodB(C aC);
    }

    alias param1_t = T;
    alias param2_t = C;

    // etc.
}

Conditional compilation

D conditional compilation is one of the unique feature of the lanquage. While in Pascal you used a custom definition like this:

a := a shr 3;
{$IFDEF MYSWITCH}
  // some conditionally compiled code
{$ENDIF}

in D the equivalent use a standard language construct version(Identifier):

a =>> 3;
version(MYSWITCH){
    // some conditionally compiled code   
}

But now you’ll get more. A feature that doesn’t exist in Pascal and Delphi is the static if expression. It allows a more advanced conditional compilation, particularly in the templates, since the type of the template parameters can be tested when it gets instantiated.

struct foo(T) {
    // something is only compiled if T type is string.
    static if (is(T == string)){
    void something(){}
    }
}

This is also related to the template constraints (the constraint is applied to the whole template while in our previous example it’s only applied to the enclosed expressions). You might know the principle if you’ve used Delphi XE6 or upper.

Inclusion

D has a feature similar to Pascal/delphi source inclusion.

for the example the following inclusion:

procedure Something;
begin
{$IFDEF DEMO}
  {$I myDemoImplementation.inc}
{$ELSE}
  {$I myFullImplementation.inc}
{$ENDIF}
end;

is rewritten in D

void something()
{
    version(DEMO)
        mixin(import(myDemoImplementation.imp));
    else
        mixin(import(myFullImplementation.imp));
}

In both cases, the file content doesn't represent a full unit/module but only some raw code which has to be compilable in the inclusion/mixin context.

RTTI, published attribute and properties.

In Pascal and Delphi, the Runtime Type Informations (RTTI) are an important feature used by the TPersistent and the TComponent system, as well as for custom serialization or for software settings or even Object dumping into a database.

D has no equivalent feature. The published attribute does not exist but instead you have some compile-time reflection (traits, std.traits, user defined attributes) which could be used to design a similar system.

Pascal properties are deeply linked to the RTTI mechanisms. D properties are a bit different and, in a way, more powerfull. They allow to use the assign operator instead of calling the setter with parens (so far this is a common behaviour) but they also can be overloaded.

A classic, non published, Pascal property (interface section only):

private
  FField: integer;
  procedure SetField(AValue: Integer);
public
  property Field: integer read FField write SetField;

The D equivalent, with overload:

private:
    int fField;
public:
    @property void field(int aValue){fField = aValue;}
    @property void field(string aValue){fField = to!int(aValue); }
    @property int field(){return fField;}

Sets and the "in" operator

In Pascal and Delphi you used the in operator for testing the presence of a value in a set. D also have this operator but its usage is different, it’s used to test the presence of a value in an associative array (AA):

auto asar = [
    8 : "AA",
    15 : "GJ",
    32 : "VJ"
];
assert(8 in asar);
assert(!(57 in asar));

The Pascal syntax used to define a set does not exist in D, for example the following expression has no simple equivalence:

const THexChars Set Of Char = ['0'..'9','a'..'f','A'..'F'];

function IsHexChar(C: Char): Boolean;
begin
  result := C in THexDigits;
end;

Additionally to the in difference, double dots .. (slice operator) are not as flexible as in Pascal. The left hand side and the right hand side must be some integral indexes.

However, as D supports operator overloading, it’s possible to extend the usage of the in operator and the slice operator in a custom type. for example:

import std.stdio;
import std.traits;

struct bitWise(T) if (isIntegral!T){
    T theValue;
    alias theValue this;
    bool opIn_r(T aValue){
        return ((aValue & theValue) >= aValue);
    }
}

void main(string args[]){
    bitWise!ubyte bw;
    bw = 0b11110000;
    assert( 0b11100000 in bw );
    assert( !(0b00001111 in bw));
}

Even better, a complete emulation of the THexChars from the previous Pascal example. The in operator is still overloaded but this time also the slice one, using two characters as arguments:

CharSet charSet;
struct CharSet
{
    private string str;
    public typeof(this) opSlice(size_t N)(char lo, char hi){
        typeof(return) result;
        foreach(c; lo .. hi+1)
            result.str ~= c;
        return result;
    } 
    public typeof(this) opIndex(Slices...)(Slices slices){
        typeof(return) result; 
        foreach(slice; slices)
            result.str ~= slice.str;
        return result;    
    } 
    public bool opIn_r(char elem){
        import std.algorithm;
        return canFind(str,elem);
    }
}

void main(string args[]){
    auto hexChars = charSet['0'..'9', 'A'..'F', 'a'..'f'];
    assert('D' in hexChars);
    assert(!('G' in hexChars));
}

The "is" operator

The keyword is exists in the two languages but have a different meaning.

In Pascal, is is used to test if a class inherits from a particular ancestor type. For example

procedure TMyClass.Event(Sender: TObject);
begin
  if (Sender is TButton) then with TButton(Sender) do begin
    // some code related to Sender as TButton...
  end;
  if (Sender is TCheckBox) then with TCheckBox(Sender) do begin
    // some code related to Sender as TCheckBox...
  end;
end;

In D, the same can be achieved by comparing the result of a cast to null:

void event(Object sender){
   Button button = cast(Button) sender;
   Checkbox checkbox = cast(Checkbox) sender;
   if(button){/*some code related to sender as Button*/}
   if(checkbox){/*some code related to sender as Checkbox */}
}

In D, is, a common usage of the is operator is to test the equivalence of two types, at compile-time. It's often used in the template constraints, for example:

template myTemplate(T) if ((is(T == float)) | (is(T == double)))
{
    T functionA(T param0, T param1){return T.init;}
    T functionB(T param0, T param1){return T.init;}
}
// compile time error, int does not verify the constraint ((is(T == float)) | (is(T == double))) 
auto value = (myTemplate!int).functionA(1,2);

Unit "Initialization" and "Finalization"

D has a similar feature to Pascal unit Initialization and unit Finalization sections. They are called static module contructors and static module destructors. Additionally to the feature similarity, there can be many of them and are executed in lexical order.

  • the Pascal way:
unit unit1;
interface
var
  Obj: TObject;
implementation
initialization
  Obj := TObject.Create;
finalization
  Obj.Free;
end.
  • the D2 way:
module module1;
Object obj;
static this(){
    obj = new Object;
}
static ~this(){
    assert(!obj); // executed at last.
}
static ~this(){
    delete obj;
}
static ~this(){
    assert(obj); // executed at first.
}

Type aliasing

Pascal has two syntaxes for aliasing a type:

type MyInt = Integer; // syntax 1
type MyOtherInt = type Integer; // syntax 2

The first syntax, which in both cases doesn't create a distinct type, is the equivalent of the following D2 declaration:

alias myInt = int;

The second syntax creates a distinct type (you may, for example, used it to create a custom property inspector). The D2 way is to use a struct combined with an alias this expression:

struct MyOtherInt {
    public int daValue;
    alias daValue this;
}

In this basic form, a MyOtherInt can be used as an int but is a distinct type. If this syntax looks a bit heavy for the few it does this is because it becomes more interesting when such a struct includes some operator overloading, like in this trivial example:

struct ExoticFloat {
    public float daValue;
    alias daValue this;
    void opAssign(string aString){
        import std.conv;
        daValue = to!float(aString);
    }
}
ExoticFloat ef;
ef = "0.1";
assert(ef == 0.1f);

Top index in loops and ranges

Pascal for loops syntax is quite different from their C-style equivalent, which includes at test and whose index variable is declared inline:

for i := 0 to high(array) do (*statement*);
for i := high(array) downto 0 do (*statement*);

is in D equivalent to:

for (auto i = 0; i < array.length; i++) /*statement*/;
for (ptrdiff_t i = array.length - 1; i > -1; i--) /*statement*/;

But the most important difference is that the top index is not included, which explains why in the reversed form, the loop index type is signed. Type inference would lead i type to be size_t (because the array property length is size_t), thus always positive and the loop wouldn't start at all.

Another syntax exists, which is less error-prone and somehow near from the Pascal for loops. The trick consists in iterating trough a range, constant or based on run-time variables. Still using the two previous examples:

foreach (i; 0 .. array.length) /*statement*/;
foreach_reverse (i; 0 .. array.length) /*statement*/;

Here again, the last element of the range is not included but the reversed form will not lead to an hidden error.

The const-ness gate

When coming from Pascal or Delphi, the const storage class for parameters makes sense for simple values, for example:

procedure something(const parameter: Integer);
var
  local_copy: Integer;
begin
  //inc(parameter); // ouch, parameter is const
  local_copy := parameter;
  inc(localcopy);
end;

In D the same is true:

void something(const int parameter);{
  //parameter++; // ouch, parameter is const
  int local_copy = parameter;
  localcopy++;
}

However, despite of the appearances, storage classes for parameters are radically different in Pascal and Delphi. In a way we could say that they are outrageously permissive, for example if you use const for a reference type, let's say a class instance:

program Project1;

type TFoo = class
  member: NativeInt;
end;

procedure FakeConst(const aFoo: TFoo);
begin
  aFoo.member := 1;
end;

var foo: TFoo;

begin
  foo := TFoo.Create;
  FakeConst(foo);
end.

Compiles fines, despite of the aFoo const-ness, while in D2, the equivalent:

module program;

class Foo{
    public size_t member;
}

void RealConst(const Foo aFoo){
    aFoo.member = 1;
}

void main(string args[]){
    auto foo = new Foo;
    RealConst(foo);
}

leads to an error:

Error: cannot modify const expression aFoo.member

because you try to modify a const parameter. D2 is more strict in regard of the const parameter storage class. For a class instance or a struct the constness is transitive.

You'll encounter the same strict behavior with modular attributes (@safe, @trusted, @system, @nogc).

Lists

In Delphi or Pascal, the classes unit provides the basic lists. In the standard D library, there is no equivalent. Why ? Because the way you used to use a list in Pascal is not efficient, for example the array syntax, classic usage of a list:

for i := 0 to list.count-1 do 
  TMySubClass(list[i]).stuff := value;

have two major problems:

list.count: a basic singly-linked-list system requires to go from first to last, so the count requires a costly traversal:

while (current <> nil) do begin
  current := current.next;
  inc(result);
end;

list[i]: for each item, an iterator starts from the beginning.

while (index <> i) do begin
  current := current.next;
  inc(index);
  result := current.data;
end;

The lists in the standard D library just give you the opportunity to add elements and to iterate with foreach(). If you want the Pascal semantic (add, remove, []...) you must wrap a list in a class, for example a Dlist.

Std lists are structs, so encapsulating one in a class is not a problem (from the memory layout point of view), their fields are accessed just like another field.

The operations you used in Delphi lists are available in std.range and std.algorithm.

Before using these modules it's necessary to return a range from the container (if you don't then you do the most common error in D2).

struct Foo{}
DList!(Foo*) lst;
foreach(i; 0 .. 10)
    lst.insertBack(new Foo);
auto usableRange = lst[];

It's also worth noting that most D code simply uses normal arrays rather than lists, and appends ~ elements directly to them, but if the content of a collection is likely to be often reorganized at run-time it's still worth using a SList or a DList.