... don't make sense. So is there a reason why they aren't excluded from the "everything in an immutable is immutable" rule?


Tomek

Hello Tomek,

> ... don't make sense. So is there a reason why they aren't excluded
> from  the "everything in an immutable is immutable" rule?
> 
> Tomek
> 

Immutable static member function could be useful...

marking struct or class as const or immutable is the same as marking every its member so. The problem I see is in definition of what const or immutable member function means:

from D specs: "Immutable member functions are guaranteed that the object and anything referred to by the this reference is immutable."

Which applies both to instance and s static functions. I think the above definition should apply only to instance functions. Static const or immutable functions should  be defined as:

Immutable static member functions are guaranteed that the static variables of object and anything referred to by these variables is immutable.

static member function cannot modify member variables already, because they do not have this pointer. They would be only allowed to modify its arguments and global variables; they would be something like pure functions for the congaing type, but not for other world.

I'm not sure how this semantics could be useful in practice, but it seems definitely better to me than erroring out on static function inside immutable type.

Dnia 07-12-2009 o 11:34:52 Michal Minich <michal.minich@gmail.com> napisał(a):

> Hello Tomek,
>
>> ... don't make sense. So is there a reason why they aren't excluded
>> from  the "everything in an immutable is immutable" rule?
>>  Tomek
>>
>
> Immutable static member function could be useful...
>
> marking struct or class as const or immutable is the same as marking every its member so. The problem I see is in definition of what const or immutable member function means:
>
> from D specs: "Immutable member functions are guaranteed that the object and anything referred to by the this reference is immutable."
>
> Which applies both to instance and s static functions. I think the above definition should apply only to instance functions. Static const or immutable functions should  be defined as:
>
> Immutable static member functions are guaranteed that the static variables of object and anything referred to by these variables is immutable.
>
> static member function cannot modify member variables already, because they do not have this pointer. They would be only allowed to modify its arguments and global variables; they would be something like pure functions for the congaing type, but not for other world.
>
> I'm not sure how this semantics could be useful in practice, but it seems definitely better to me than erroring out on static function inside immutable type.
>

If by the lot of text above you meant it's a compiler bug, I agree :-)
http://d.puremagic.com/issues/show_bug.cgi?id=3598


Tomek

Hello Tomek,

> Dnia 07-12-2009 o 11:34:52 Michal Minich <michal.minich@gmail.com>
> napisał(a):
> 
>> Hello Tomek,
>> 
>>> ... don't make sense. So is there a reason why they aren't excluded
>>> from  the "everything in an immutable is immutable" rule?
>>> Tomek
>> Immutable static member function could be useful...
>> 
>> marking struct or class as const or immutable is the same as marking
>> every its member so. The problem I see is in definition of what const
>> or  immutable member function means:
>> 
>> from D specs: "Immutable member functions are guaranteed that the
>> object  and anything referred to by the this reference is immutable."
>> 
>> Which applies both to instance and s static functions. I think the
>> above  definition should apply only to instance functions. Static
>> const or  immutable functions should  be defined as:
>> 
>> Immutable static member functions are guaranteed that the static
>> variables of object and anything referred to by these variables is
>> immutable.
>> 
>> static member function cannot modify member variables already,
>> because  they do not have this pointer. They would be only allowed to
>> modify its  arguments and global variables; they would be something
>> like pure  functions for the congaing type, but not for other world.
>> 
>> I'm not sure how this semantics could be useful in practice, but it
>> seems definitely better to me than erroring out on static function
>> inside immutable type.
>> 
> If by the lot of text above you meant it's a compiler bug, I agree :-)
> http://d.puremagic.com/issues/show_bug.cgi?id=3598
> 
> Tomek
> 

I'm glad we agreed it is a bug :-)

How about the semantics I proposed ? If you agree I would include it in the bug text.

On Mon, 07 Dec 2009 05:34:52 -0500, Michal Minich <michal.minich@gmail.com> wrote:

> Hello Tomek,
>
>> ... don't make sense. So is there a reason why they aren't excluded
>> from  the "everything in an immutable is immutable" rule?
>>  Tomek
>>
>
> Immutable static member function could be useful...
>
> marking struct or class as const or immutable is the same as marking every its member so. The problem I see is in definition of what const or immutable member function means:
>
> from D specs: "Immutable member functions are guaranteed that the object and anything referred to by the this reference is immutable."
>
> Which applies both to instance and s static functions. I think the above definition should apply only to instance functions. Static const or immutable functions should  be defined as:
>
> Immutable static member functions are guaranteed that the static variables of object and anything referred to by these variables is immutable.
>

Let's say there was an immutable static member function which treated the static data members as immutable.  This means nobody else can change them.  This means that there cannot be any static member functions that are not immutable.  This means that the data members cannot be marked as public unless they are also marked as immutable.

In this case, the equivalent thing is to mark all static data members as immutable, and leave the static function as a normal function.

This is why the OP said they "don't make sense"  Although his statement really should be "immutable static functions don't make sense", not "static functions in immutable types don't make sense," as they do make sense, they should just not be marked as immutable.

The bug filed is absolutely a bug, it should be possible to have a normal static function inside an immutable type.

-Steve

Hello Steven,

I'm sorry, but have big trouble to understand what you want to say. So I will comment to what I can, and try to rephrase my post.

> On Mon, 07 Dec 2009 05:34:52 -0500, Michal Minich
> <michal.minich@gmail.com> wrote:
> 
>> Hello Tomek,
>> 
>>> ... don't make sense. So is there a reason why they aren't excluded
>>> from  the "everything in an immutable is immutable" rule?
>>> Tomek
>> Immutable static member function could be useful...
>> 
>> marking struct or class as const or immutable is the same as marking
>> every its member so. The problem I see is in definition of what const
>> or  immutable member function means:
>> 
>> from D specs: "Immutable member functions are guaranteed that the
>> object  and anything referred to by the this reference is immutable."
>> 
>> Which applies both to instance and s static functions. I think the
>> above  definition should apply only to instance functions. Static
>> const or  immutable functions should  be defined as:
>> 
>> Immutable static member functions are guaranteed that the static
>> variables of object and anything referred to by these variables is
>> immutable.
>> 
> Let's say there was an immutable static member function which treated
> the  static data members as immutable.  This means nobody else can
> change  them.  

Couldn't they be changed from outside of the type?

> This means that there cannot be any static member
> functions that  are not immutable.  This means that the data members
> cannot be marked as  public unless they are also marked as immutable.
> 
> In this case, the equivalent thing is to mark all static data members
> as  immutable, and leave the static function as a normal function.
> 
> This is why the OP said they "don't make sense"  Although his
> statement  really should be "immutable static functions don't make
> sense", not  "static functions in immutable types don't make sense,"
> as they do make  sense, they should just not be marked as immutable.
> 
> The bug filed is absolutely a bug, it should be possible to have a
> normal  static function inside an immutable type.

Currently everything is marked as immutable in immutable type. Do you think it is good to have exception for static function?
> 
> -Steve
> 

So to rephrase:
1. from D specs: " A struct declaration can have a storage class of const, immutable or shared. It has an equivalent effect as declaring each member of the struct as const, immutable or shared." -- I'm ok with this, it is simple.

2. from D specs: "Immutable member functions are guaranteed that the object and anything referred to by the this reference is immutable." -- this rule is wrong, because it mentions *this* and at the same time it applies to static member functions, which obviously doesn't have *this* reference.

3. I propose changing rule in point 2 to: "Immutable *non-static* member functions are guaranteed that the object and anything referred to by the this reference is immutable." and adding this one: "Immutable static member functions are guaranteed that the static variables of object and anything referred to by these variables is immutable"

And I'm asking if the change is reasonable - Or how should be defined semantics of static immutable member function?

On Tue, 08 Dec 2009 11:53:03 -0500, Michal Minich <michal.minich@gmail.com> wrote:

> Hello Steven,
>
> I'm sorry, but have big trouble to understand what you want to say. So I will comment to what I can, and try to rephrase my post.
>
>> On Mon, 07 Dec 2009 05:34:52 -0500, Michal Minich
>> <michal.minich@gmail.com> wrote:
>>
>>> Hello Tomek,
>>>
>>>> ... don't make sense. So is there a reason why they aren't excluded
>>>> from  the "everything in an immutable is immutable" rule?
>>>> Tomek
>>> Immutable static member function could be useful...
>>>  marking struct or class as const or immutable is the same as marking
>>> every its member so. The problem I see is in definition of what const
>>> or  immutable member function means:
>>>  from D specs: "Immutable member functions are guaranteed that the
>>> object  and anything referred to by the this reference is immutable."
>>>  Which applies both to instance and s static functions. I think the
>>> above  definition should apply only to instance functions. Static
>>> const or  immutable functions should  be defined as:
>>>  Immutable static member functions are guaranteed that the static
>>> variables of object and anything referred to by these variables is
>>> immutable.
>>>
>> Let's say there was an immutable static member function which treated
>> the  static data members as immutable.  This means nobody else can
>> change  them.
>
> Couldn't they be changed from outside of the type?

That would be a violation of immutability.  If they can change, then they are not immutable.  Note they cannot be "temporarily" immutable, they are immutable from the moment that something casts it to immutable.  It is up to you, the programmer, to ensure that no mutable references exist after you perform the cast.  Using a mutable reference to immutable data results in undefined behavior.

To illustrate:

int x;
immutable(int) *y;

void foo()
{
   y = cast(immutable(int)*)&y;  // no longer allowed to use x!
}

void bar()
{
   foo();
   x++; // undefined behavior!
}

>
>> This means that there cannot be any static member
>> functions that  are not immutable.  This means that the data members
>> cannot be marked as  public unless they are also marked as immutable.
>>  In this case, the equivalent thing is to mark all static data members
>> as  immutable, and leave the static function as a normal function.
>>  This is why the OP said they "don't make sense"  Although his
>> statement  really should be "immutable static functions don't make
>> sense", not  "static functions in immutable types don't make sense,"
>> as they do make  sense, they should just not be marked as immutable.
>>  The bug filed is absolutely a bug, it should be possible to have a
>> normal  static function inside an immutable type.
>
> Currently everything is marked as immutable in immutable type. Do you think it is good to have exception for static function?

Either you make an exception for a static function -or- applying immutable to a static function is a no-op.  There is no reason to restrict immutable types from having static functions.

> So to rephrase:
> 1. from D specs: " A struct declaration can have a storage class of const, immutable or shared. It has an equivalent effect as declaring each member of the struct as const, immutable or shared." -- I'm ok with this, it is simple.
>
> 2. from D specs: "Immutable member functions are guaranteed that the object and anything referred to by the this reference is immutable." -- this rule is wrong, because it mentions *this* and at the same time it applies to static member functions, which obviously doesn't have *this* reference.
>
> 3. I propose changing rule in point 2 to: "Immutable *non-static* member functions are guaranteed that the object and anything referred to by the this reference is immutable." and adding this one: "Immutable static member functions are guaranteed that the static variables of object and anything referred to by these variables is immutable"
>
> And I'm asking if the change is reasonable - Or how should be defined semantics of static immutable member function?

static immutable member functions are either illegal in the case where all static member data is not immutable or equivalent to static member functions in the case where all the member data is immutable.

I agree the rule needs to change, here is how I would change it:

A struct declaration can have a storage class of const, immutable or shared. It has an equivalent effect as declaring each member of the struct except static functions as const, immutable or shared.

Immutable/const/shared non-static member functions are guaranteed that the object and anything referred to by the this reference is immutable/const/shared.  Declaring a static member function is immutable/const/shared is considered an error.

-Steve

On Tue, 08 Dec 2009 12:23:53 -0500, Steven Schveighoffer wrote:


> That would be a violation of immutability.  If they can change, then
> they are not immutable.  Note they cannot be "temporarily" immutable,
> they are immutable from the moment that something casts it to immutable.
>  It is up to you, the programmer, to ensure that no mutable references
> exist after you perform the cast.  Using a mutable reference to
> immutable data results in undefined behavior.
> 
> To illustrate:
> 
> int x;
> immutable(int) *y;
> 
> void foo()
> {
>     y = cast(immutable(int)*)&y;  // no longer allowed to use x!
> }
> 
> void bar()
> {
>     foo();
>     x++; // undefined behavior!
> }
> 
This is misunderstanding, I'm talking about interaction of mutable static data members with immutable static member function, as in this example:

struct S
{
    static int x;

    static void foo () immutable
    {
       // x = 3; // should be error, because immutable static member
function cannot change mutable static data members.
    }
}

> 
>>> This means that there cannot be any static member functions that  are
>>> not immutable.  This means that the data members cannot be marked as
>>> public unless they are also marked as immutable.
>>>  In this case, the equivalent thing is to mark all static data members
>>> as  immutable, and leave the static function as a normal function.
>>>  This is why the OP said they "don't make sense"  Although his
>>> statement  really should be "immutable static functions don't make
>>> sense", not  "static functions in immutable types don't make sense,"
>>> as they do make  sense, they should just not be marked as immutable.
>>>  The bug filed is absolutely a bug, it should be possible to have a
>>> normal  static function inside an immutable type.
>>
>> Currently everything is marked as immutable in immutable type. Do you think it is good to have exception for static function?
> 
> Either you make an exception for a static function -or- applying immutable to a static function is a no-op.  There is no reason to restrict immutable types from having static functions.
> 
>> So to rephrase:
>> 1. from D specs: " A struct declaration can have a storage class of
>> const, immutable or shared. It has an equivalent effect as declaring
>> each member of the struct as const, immutable or shared." -- I'm ok
>> with this, it is simple.
>>
>> 2. from D specs: "Immutable member functions are guaranteed that the object and anything referred to by the this reference is immutable." -- this rule is wrong, because it mentions *this* and at the same time it applies to static member functions, which obviously doesn't have *this* reference.
>>
>> 3. I propose changing rule in point 2 to: "Immutable *non-static* member functions are guaranteed that the object and anything referred to by the this reference is immutable." and adding this one: "Immutable static member functions are guaranteed that the static variables of object and anything referred to by these variables is immutable"
>>
>> And I'm asking if the change is reasonable - Or how should be defined semantics of static immutable member function?
> 
> static immutable member functions are either illegal in the case where all static member data is not immutable or equivalent to static member functions in the case where all the member data is immutable.
> 
> I agree the rule needs to change, here is how I would change it:
> 
> A struct declaration can have a storage class of const, immutable or shared. It has an equivalent effect as declaring each member of the struct except static functions as const, immutable or shared.
> 
> Immutable/const/shared non-static member functions are guaranteed that the object and anything referred to by the this reference is immutable/const/shared.  Declaring a static member function is immutable/const/shared is considered an error.

The change I proposed is in at least line with existing functionality. What is a benefit of not having immutable member functions as you proposed?

> 
> -Steve

Moved the discussion to digitalmars.D group. please reply there