www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Neat trick - 'with' and unnamed objects + 'with' proposals

reply Bill Baxter <dnewsgroup billbaxter.com> writes:
I just discovered this little trick that I think is way cool.
You can use 'with' in combination with an anonymous object to set 
attributes on that object without having to bother giving it a name or a 
variable.


with(new someGUIWidget(myParent)) {
    shown = true;
    focusable = false;
}

That's really nifty for GUI code.

This could be even more useful if there were some kind of 'this' for 
'with' blocks.  Then you could also call non-member functions.

with(new someGUIWidget(myParent)) {
    shown = true;
    focusable = false;
    some_global_function(this);
}

Or perhaps analogous to how constructors are called 'this()' and have a 
'this' pointer, we'd have a 'with' pointer so that the meaning of 'this' 
in a class wouldn't be shadowed:

with(new someGUIWidget(myParent)) {
    some_global_function(with);
}

And even cooler would be if you could have a 'with-expression' that just 
evaluates to the thing in the parens:

auto obj = with(new someGUIWidget(myParent))
{
    shown = true;
    focusable = false;
    some_global_function(this);
};


Sometimes I've seen people try to make APIs in C++ where all the 
attribute setters return the object so that property setting can be 
chained, like:

      myObject.set_focusable(true).set_shown(true);

Presumably the objective is to avoid repeating the name of the object a 
lot.  But that never works out too well, because making every mutator 
return a pointer to 'this' is just not so practical in the end, and 
forcing every property setting operation to use function call syntax is 
also not so great.

--bb
Apr 23 2007
next sibling parent reply "Denis Golovan" <denis remove.me.azovprom.com> writes:
 Or perhaps analogous to how constructors are called 'this()' and have a
 'this' pointer, we'd have a 'with' pointer so that the meaning of 'this' 
 in a class wouldn't be shadowed:


  +1. That feature is useful sometimes.


 --bb
Apr 24 2007
parent reply Bill Baxter <dnewsgroup billbaxter.com> writes:
Denis Golovan wrote:

 Or perhaps analogous to how constructors are called 'this()' and have a
 'this' pointer, we'd have a 'with' pointer so that the meaning of 'this' 
 in a class wouldn't be shadowed:


 
   +1. That feature is useful sometimes.


You say "is" as if it exists in some other language that you've used? 
Does it? And if so what's the syntax?

--bb
Apr 24 2007
next sibling parent Max Samukha <samukha voliacable.com> writes:
On Tue, 24 Apr 2007 22:03:07 +0900, Bill Baxter
<dnewsgroup billbaxter.com> wrote:


Denis Golovan wrote:

 Or perhaps analogous to how constructors are called 'this()' and have a
 'this' pointer, we'd have a 'with' pointer so that the meaning of 'this' 
 in a class wouldn't be shadowed:


 
   +1. That feature is useful sometimes.


You say "is" as if it exists in some other language that you've used? 
Does it? And if so what's the syntax?

--bb



don't provide a way to call global (and local, I guess) functions with
'this'.
http://community.bartdesmet.net/blogs/bart/archive/2006/12/04/C_2300_-3.0-Feature-Focus-_2D00_-Part-2-_2D00_-Object-Initializers.aspx
Apr 24 2007
prev sibling parent eao197 <eao197 intervale.ru> writes:
On Tue, 24 Apr 2007 17:03:07 +0400, Bill Baxter  =

<dnewsgroup billbaxter.com> wrote:


 Denis Golovan wrote:

 Or perhaps analogous to how constructors are called 'this()' and hav=






e a

 'this' pointer, we'd have a 'with' pointer so that the meaning of  =







 'this' in a class wouldn't be shadowed:


    +1. That feature is useful sometimes.


 You say "is" as if it exists in some other language that you've used? =


 =


 Does it? And if so what's the syntax?


Such behaviour can easily be implemented in Ruby. This is very simple  =

demonstration:


   class Demo
     attr_reader :shown
     attr_reader :focused
   end


   class AnotherDemo
     attr_reader :title
   end


   def some_global_function( obj )
     p obj
   end


   def with( obj, &blk )

     class <<obj
       def with; self; end
     end




     begin
       obj.instance_eval &blk
     ensure

       class <<obj
         remove_method( :with )
       end
     end

     obj
   end


   d =3D with( Demo.new ) {
      shown =3D false
      focused =3D true

     some_global_function( with )
   }
   a =3D with( AnotherDemo.new ) {
      title =3D 'Hello, World'

     some_global_function( with )
   }





That program prints:



d: shown: false, focused: true
a: title: Hello, World

Such implementation of 'with' assumes that obj hasn't method 'with'. But=
  =

this limitation can be removed via aliasing.

-- =

Regards,
Yauheni Akhotnikau
Apr 24 2007
prev sibling next sibling parent reply Dan <murpsoft hotmail.com> writes:
Bill Baxter Wrote:


 I just discovered this little trick that I think is way cool.
 You can use 'with' in combination with an anonymous object to set 
 attributes on that object without having to bother giving it a name or a 
 variable.
 
 
 with(new someGUIWidget(myParent)) {
     shown = true;
     focusable = false;
 }


In JavaScript they've had this "feature" for a very long time.  I never, ever
use it.  Why?

x = 3;
myFunc = function(){
x = 2;
this.x = 1;
bla bla bla ginger
with(x){
bla bla bla bla ginger
bla bla bla
bla bla
x += 3;
bla bla
}
x -= 2;
}

That in itself isn't so bad.  But in real code, it looks godawful and makes a
guy concentrate on scoping it for a second; distracting him from the algorithm.


 That's really nifty for GUI code.
 
 This could be even more useful if there were some kind of 'this' for 
 'with' blocks.  Then you could also call non-member functions.


Lions and tigers and bears, oh my.  Can't you just:

alias supercalafragilisticexpialadocious.bob.mom.mom.niece.daughter d;

d.shown = true;
d.focusable = false;
etc.?


 with(new someGUIWidget(myParent)) {
     shown = true;
     focusable = false;
     some_global_function(this);
 }
 
 Or perhaps analogous to how constructors are called 'this()' and have a 
 'this' pointer, we'd have a 'with' pointer so that the meaning of 'this' 
 in a class wouldn't be shadowed:
 
 with(new someGUIWidget(myParent)) {
     some_global_function(with);
 }


Eww... ambiguity and context sensitivity.


 And even cooler would be if you could have a 'with-expression' that just 
 evaluates to the thing in the parens:
 
 auto obj = with(new someGUIWidget(myParent))
 {
     shown = true;
     focusable = false;
     some_global_function(this);
 };


auto obj = new someGuiWidget(myParent){
  shown = true;
  focusable = false;
};
some_global_function(obj); <- can't execute in a class declaration.  Try in the
constructor?


 
 Sometimes I've seen people try to make APIs in C++ where all the 
 attribute setters return the object so that property setting can be 
 chained, like:
 
       myObject.set_focusable(true).set_shown(true);


I did that back in the day too, but not for setting properties, it was for
performing more complex algorithms with the DOM, such as making an object move
in circles, or fall using a gravity formula.


 Presumably the objective is to avoid repeating the name of the object a 
 lot.  But that never works out too well, because making every mutator 
 return a pointer to 'this' is just not so practical in the end, and 
 forcing every property setting operation to use function call syntax is 
 also not so great.


Property setting operations?  Totally agree.  Returning a this* isn't so bad
unless you have something else to return.  I only did it with subroutines.

The real reason it's bad is because you're masturbating the stack doing calls
like that just to avoid doing something in source code.  Such a thing is
forgiveable in web scripts where filesize matters, it's not in programming.
Apr 24 2007
parent Bill Baxter <dnewsgroup billbaxter.com> writes:
Dan wrote:

 Bill Baxter Wrote:
 

 I just discovered this little trick that I think is way cool.
 You can use 'with' in combination with an anonymous object to set 
 attributes on that object without having to bother giving it a name or a 
 variable.


 with(new someGUIWidget(myParent)) {
     shown = true;
     focusable = false;
 }


 
 In JavaScript they've had this "feature" for a very long time.  I never, ever
use it.  Why?
 
 x = 3;
 myFunc = function(){
 x = 2;
 this.x = 1;
 bla bla bla ginger
 with(x){
 bla bla bla bla ginger
 bla bla bla
 bla bla
 x += 3;
 bla bla
 }
 x -= 2;
 }
 
 That in itself isn't so bad.  But in real code, it looks godawful and makes a
guy concentrate on scoping it for a second; distracting him from the algorithm.



You seem to be harping on 'with' in general, not with what I was 
actually talking about which is using it on unnamed variables.  'with' 
itself has been in D as long as I've known D, I just didn't realize you 
could use it with anonymous objects.

Anyway, you certainly don't have to like it or use it.  But I find it 
useful in moderation.

--bb
Apr 24 2007
prev sibling parent reply BCS <BCS pathlink.com> writes:
Bill Baxter wrote:

 I just discovered this little trick that I think is way cool.
 You can use 'with' in combination with an anonymous object to set 
 attributes on that object without having to bother giving it a name or a 
 variable.
 
 
 with(new someGUIWidget(myParent)) {
    shown = true;
    focusable = false;
 }
 
 That's really nifty for GUI code.
 
 This could be even more useful if there were some kind of 'this' for 
 'with' blocks.  Then you could also call non-member functions.
 
 with(new someGUIWidget(myParent)) {
    shown = true;
    focusable = false;
    some_global_function(this);
 }
 
 Or perhaps analogous to how constructors are called 'this()' and have a 
 'this' pointer, we'd have a 'with' pointer so that the meaning of 'this' 
 in a class wouldn't be shadowed:
 
 with(new someGUIWidget(myParent)) {
    some_global_function(with);
 }
 
 And even cooler would be if you could have a 'with-expression' that just 
 evaluates to the thing in the parens:
 
 auto obj = with(new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(this);
 };
 
 
 Sometimes I've seen people try to make APIs in C++ where all the 
 attribute setters return the object so that property setting can be 
 chained, like:
 
      myObject.set_focusable(true).set_shown(true);
 
 Presumably the objective is to avoid repeating the name of the object a 
 lot.  But that never works out too well, because making every mutator 
 return a pointer to 'this' is just not so practical in the end, and 
 forcing every property setting operation to use function call syntax is 
 also not so great.
 
 --bb



how about take a trick from "if"

with(auto name = new someGUIWidget(myParent))
{
    shown = true;
    focusable = false;
    some_global_function(name);
}
Apr 24 2007
parent reply Bill Baxter <dnewsgroup billbaxter.com> writes:
BCS wrote:

 Bill Baxter wrote:



 how about take a trick from "if"
 
 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }


Yeh, I was actually just thinking that myself, and was disappointed to 
find out it doesn't work.  :-(

Looking at the spec it seems that's special to 'if'.   It could be 
useful for switch(), with() and synchronized(), but it seems it hasn't 
been added.

--bb
Apr 24 2007
parent reply BCS <BCS pathlink.com> writes:
Bill Baxter wrote:

 BCS wrote:
 

 Bill Baxter wrote:


 
 

 how about take a trick from "if"

 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }


 
 
 Yeh, I was actually just thinking that myself, and was disappointed to 
 find out it doesn't work.  :-(
 
 Looking at the spec it seems that's special to 'if'.   It could be 
 useful for switch(), with() and synchronized(), but it seems it hasn't 
 been added.
 
 --bb


Ohhh. That would be nice in switch ... and while...
Apr 24 2007
parent reply Bill Baxter <dnewsgroup billbaxter.com> writes:
BCS wrote:

 Bill Baxter wrote:

 BCS wrote:


 Bill Baxter wrote:




 how about take a trick from "if"

 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }



 Yeh, I was actually just thinking that myself, and was disappointed to 
 find out it doesn't work.  :-(

 Looking at the spec it seems that's special to 'if'.   It could be 
 useful for switch(), with() and synchronized(), but it seems it hasn't 
 been added.

 --bb


 
 Ohhh. That would be nice in switch ... and while...


I was thinking 'while' would have issues, because the condition is 
evaluated multiple times.  So I left it off the list.  Maybe it's not a 
problem though.  Just make it equiv to
{
    typeof(condition()) x;
    while(x = condition()) {
        // do stuff
    }
}

--bb
Apr 24 2007
next sibling parent reply Max Samukha <samukha voliacable.com> writes:
On Wed, 25 Apr 2007 01:53:56 +0900, Bill Baxter
<dnewsgroup billbaxter.com> wrote:


BCS wrote:

 Bill Baxter wrote:

 BCS wrote:


 Bill Baxter wrote:




 how about take a trick from "if"

 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }



 Yeh, I was actually just thinking that myself, and was disappointed to 
 find out it doesn't work.  :-(

 Looking at the spec it seems that's special to 'if'.   It could be 
 useful for switch(), with() and synchronized(), but it seems it hasn't 
 been added.

 --bb


 
 Ohhh. That would be nice in switch ... and while...


I was thinking 'while' would have issues, because the condition is 
evaluated multiple times.  So I left it off the list.  Maybe it's not a 
problem though.  Just make it equiv to
{
    typeof(condition()) x;
    while(x = condition()) {
        // do stuff
    }
}

--bb


This would be inconsistent with if(), which declares the variable to
be local to if's scope. BTW, it'd be nice to have the variable
declared outside the if block as you proposed for 'with'. I remember a
case or two when I wanted that.
Apr 24 2007
next sibling parent reply Max Samukha <samukha voliacable.com> writes:
On Tue, 24 Apr 2007 20:18:24 +0300, Max Samukha
<samukha voliacable.com> wrote:


On Wed, 25 Apr 2007 01:53:56 +0900, Bill Baxter
<dnewsgroup billbaxter.com> wrote:


BCS wrote:

 Bill Baxter wrote:

 BCS wrote:


 Bill Baxter wrote:




 how about take a trick from "if"

 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }



 Yeh, I was actually just thinking that myself, and was disappointed to 
 find out it doesn't work.  :-(

 Looking at the spec it seems that's special to 'if'.   It could be 
 useful for switch(), with() and synchronized(), but it seems it hasn't 
 been added.

 --bb


 
 Ohhh. That would be nice in switch ... and while...


I was thinking 'while' would have issues, because the condition is 
evaluated multiple times.  So I left it off the list.  Maybe it's not a 
problem though.  Just make it equiv to
{
    typeof(condition()) x;
    while(x = condition()) {
        // do stuff
    }
}

--bb


This would be inconsistent with if(), which declares the variable to
be local to if's scope. BTW, it'd be nice to have the variable
declared outside the if block as you proposed for 'with'. I remember a
case or two when I wanted that.


 


BTW, there is a bug

void main(char[][] f)
{
    int x = 1;

    if (auto x = 2)
    {
        writefln(x);
    }

    writefln(x);
}

x in if() shadows the outer x and it shouldn't according to the specs
Apr 24 2007
parent reply Bill Baxter <dnewsgroup billbaxter.com> writes:
Max Samukha wrote:

 On Tue, 24 Apr 2007 20:18:24 +0300, Max Samukha
 <samukha voliacable.com> wrote:
 

 On Wed, 25 Apr 2007 01:53:56 +0900, Bill Baxter
 <dnewsgroup billbaxter.com> wrote:


 BCS wrote:

 Bill Baxter wrote:

 BCS wrote:


 Bill Baxter wrote:



 how about take a trick from "if"

 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }


 Yeh, I was actually just thinking that myself, and was disappointed to 
 find out it doesn't work.  :-(

 Looking at the spec it seems that's special to 'if'.   It could be 
 useful for switch(), with() and synchronized(), but it seems it hasn't 
 been added.

 --bb


 Ohhh. That would be nice in switch ... and while...


 I was thinking 'while' would have issues, because the condition is 
 evaluated multiple times.  So I left it off the list.  Maybe it's not a 
 problem though.  Just make it equiv to
 {
    typeof(condition()) x;
    while(x = condition()) {
        // do stuff
    }
 }

 --bb


 This would be inconsistent with if(), which declares the variable to
 be local to if's scope. BTW, it'd be nice to have the variable
 declared outside the if block as you proposed for 'with'. I remember a
 case or two when I wanted that.


 
 BTW, there is a bug
 
 void main(char[][] f)
 {
     int x = 1;
 
     if (auto x = 2)
     {
         writefln(x);
     }
 
     writefln(x);
 }
 
 x in if() shadows the outer x and it shouldn't according to the specs


You mean it should be an error according to the specs, right?
--bb
Apr 24 2007
parent Max Samukha <samukha voliacable.com> writes:
On Wed, 25 Apr 2007 02:38:26 +0900, Bill Baxter
<dnewsgroup billbaxter.com> wrote:


Max Samukha wrote:

 On Tue, 24 Apr 2007 20:18:24 +0300, Max Samukha
 <samukha voliacable.com> wrote:
 

 On Wed, 25 Apr 2007 01:53:56 +0900, Bill Baxter
 <dnewsgroup billbaxter.com> wrote:


 BCS wrote:

 Bill Baxter wrote:

 BCS wrote:


 Bill Baxter wrote:



 how about take a trick from "if"

 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }


 Yeh, I was actually just thinking that myself, and was disappointed to 
 find out it doesn't work.  :-(

 Looking at the spec it seems that's special to 'if'.   It could be 
 useful for switch(), with() and synchronized(), but it seems it hasn't 
 been added.

 --bb


 Ohhh. That would be nice in switch ... and while...


 I was thinking 'while' would have issues, because the condition is 
 evaluated multiple times.  So I left it off the list.  Maybe it's not a 
 problem though.  Just make it equiv to
 {
    typeof(condition()) x;
    while(x = condition()) {
        // do stuff
    }
 }

 --bb


 This would be inconsistent with if(), which declares the variable to
 be local to if's scope. BTW, it'd be nice to have the variable
 declared outside the if block as you proposed for 'with'. I remember a
 case or two when I wanted that.


 
 BTW, there is a bug
 
 void main(char[][] f)
 {
     int x = 1;
 
     if (auto x = 2)
     {
         writefln(x);
     }
 
     writefln(x);
 }
 
 x in if() shadows the outer x and it shouldn't according to the specs


You mean it should be an error according to the specs, right?
--bb


Yes
Apr 24 2007
prev sibling parent reply Bill Baxter <dnewsgroup billbaxter.com> writes:
Max Samukha wrote:

 On Wed, 25 Apr 2007 01:53:56 +0900, Bill Baxter
 <dnewsgroup billbaxter.com> wrote:
 

 BCS wrote:

 Bill Baxter wrote:

 BCS wrote:


 Bill Baxter wrote:



 how about take a trick from "if"

 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }


 Yeh, I was actually just thinking that myself, and was disappointed to 
 find out it doesn't work.  :-(

 Looking at the spec it seems that's special to 'if'.   It could be 
 useful for switch(), with() and synchronized(), but it seems it hasn't 
 been added.

 --bb


 Ohhh. That would be nice in switch ... and while...


 I was thinking 'while' would have issues, because the condition is 
 evaluated multiple times.  So I left it off the list.  Maybe it's not a 
 problem though.  Just make it equiv to
 {
    typeof(condition()) x;
    while(x = condition()) {
        // do stuff
    }
 }

 --bb


 
 This would be inconsistent with if(), which declares the variable to
 be local to if's scope. BTW, it'd be nice to have the variable
 declared outside the if block as you proposed for 'with'. I remember a
 case or two when I wanted that.


No that's not what I meant.  Note the extra block scope _outside_ the 
while.  That would be silently introduced by the transformation.  The 
scope added is a scope that you have no way of interacting with or 
introducing code into, so effectively it *is* limiting the var scope to 
the scope of the while().  Of course there are no "scopes" anyway after 
everything gets compiled down to ASM, so its all just pleasant fictions 
to help us poor humans understand what's going on.

--bb
Apr 24 2007
parent Max Samukha <samukha voliacable.com> writes:
On Wed, 25 Apr 2007 02:35:02 +0900, Bill Baxter
<dnewsgroup billbaxter.com> wrote:


Max Samukha wrote:

 On Wed, 25 Apr 2007 01:53:56 +0900, Bill Baxter
 <dnewsgroup billbaxter.com> wrote:
 

 BCS wrote:

 Bill Baxter wrote:

 BCS wrote:


 Bill Baxter wrote:



 how about take a trick from "if"

 with(auto name = new someGUIWidget(myParent))
 {
    shown = true;
    focusable = false;
    some_global_function(name);
 }


 Yeh, I was actually just thinking that myself, and was disappointed to 
 find out it doesn't work.  :-(

 Looking at the spec it seems that's special to 'if'.   It could be 
 useful for switch(), with() and synchronized(), but it seems it hasn't 
 been added.

 --bb


 Ohhh. That would be nice in switch ... and while...


 I was thinking 'while' would have issues, because the condition is 
 evaluated multiple times.  So I left it off the list.  Maybe it's not a 
 problem though.  Just make it equiv to
 {
    typeof(condition()) x;
    while(x = condition()) {
        // do stuff
    }
 }

 --bb


 
 This would be inconsistent with if(), which declares the variable to
 be local to if's scope. BTW, it'd be nice to have the variable
 declared outside the if block as you proposed for 'with'. I remember a
 case or two when I wanted that.


No that's not what I meant.  Note the extra block scope _outside_ the 
while.  That would be silently introduced by the transformation.  The 
scope added is a scope that you have no way of interacting with or 
introducing code into, so effectively it *is* limiting the var scope to 
the scope of the while().  Of course there are no "scopes" anyway after 
everything gets compiled down to ASM, so its all just pleasant fictions 
to help us poor humans understand what's going on.

--bb


 I understand now. But in your second proposal you want the object to
be accessible outside the 'with' block, so would it better to declare
the variable outside the scope in the first place? 

with(auto name = new someGUIWidget(myParent))
{
  shown = true;
  focusable = false;
  some_global_function(name);
}

name.shown = false;
Apr 24 2007
prev sibling parent reply Dan <murpsoft hotmail.com> writes:
Bill Baxter Wrote:

     typeof(condition()) x;
     while(x = condition()) {
         // do stuff
     }
 }


Which is frighteningly different semantically, while being almost identical
syntactically to:

while(x == condition()) {

Which is probably why Walter didn't want it to work.  It concerns me a little
that if(x = 3) works.  

Honestly, D is syntactically abstracted away enough that while(something
something) wouldn't have to compile down to evaluating it repeatedly.
Apr 24 2007
parent BCS <BCS pathlink.com> writes:
Dan wrote:

 Bill Baxter Wrote:
 

    typeof(condition()) x;
    while(x = condition()) {
        // do stuff
    }
}


 
 
 Which is frighteningly different semantically, while being almost identical
syntactically to:
 
 while(x == condition()) {
 
 Which is probably why Walter didn't want it to work.  It concerns me a little
that if(x = 3) works.  
 
 Honestly, D is syntactically abstracted away enough that while(something
something) wouldn't have to compile down to evaluating it repeatedly.
 
 


if(x = 3) doesn't compile. For the If case the syntax requiters that the 
type specifier be used:

if(auto x = 3) // works
if(x == 3) // works
if(x = 3) // fails

OTOH with 'wile' the third case is totally legitimate

char x;
while(x = getc()) // scan for null and use non null in loop;

actually the proposed addition would make that error less likely, in 
fact the direct assignment could be banned in a while statement.

char x;

while(auto c = getc())
{
	x=c; // if last c needed;
	break;
}
Apr 24 2007