www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.announce - =?UTF-8?B?U21hcnRSZWbvvJo=?= The Smart Pointer In D

reply Dsby <dushibaiyu yahoo.com> writes:
I write the ref count pointer and the scoped point in D.
it just Like cpp's shared_ptr , waek_ptr and unique_ptr .
Now, it is  Developing.
I will write more test before the frist release.
And the docs is null.
It on github: https://github.com/huntlabs/SmartRef
Jan 13
next sibling parent reply nbro <bro gmail.com> writes:
On Friday, 13 January 2017 at 16:50:37 UTC, Dsby wrote:
 I write the ref count pointer and the scoped point in D.
 it just Like cpp's shared_ptr , waek_ptr and unique_ptr .
 Now, it is  Developing.
 I will write more test before the frist release.
 And the docs is null.
 It on github: https://github.com/huntlabs/SmartRef
What's would be the advantages of smart pointers in D?
Jan 13
next sibling parent reply Swoorup Joshi <swoorupjoshi gmail.com> writes:
On Saturday, 14 January 2017 at 02:05:11 UTC, nbro wrote:
 On Friday, 13 January 2017 at 16:50:37 UTC, Dsby wrote:
 I write the ref count pointer and the scoped point in D.
 it just Like cpp's shared_ptr , waek_ptr and unique_ptr .
 Now, it is  Developing.
 I will write more test before the frist release.
 And the docs is null.
 It on github: https://github.com/huntlabs/SmartRef
What's would be the advantages of smart pointers in D?
Simple, Same Advantages you would get with C++ smart pointers.
Jan 13
parent reply Dmitry Olshansky <dmitry.olsh gmail.com> writes:
On 1/14/17 3:20 AM, Swoorup Joshi wrote:
 On Saturday, 14 January 2017 at 02:05:11 UTC, nbro wrote:
 On Friday, 13 January 2017 at 16:50:37 UTC, Dsby wrote:
 I write the ref count pointer and the scoped point in D.
 it just Like cpp's shared_ptr , waek_ptr and unique_ptr .
 Now, it is  Developing.
 I will write more test before the frist release.
 And the docs is null.
 It on github: https://github.com/huntlabs/SmartRef
What's would be the advantages of smart pointers in D?
Simple, Same Advantages you would get with C++ smart pointers.
Not quite true, there is at least one huge advantage - thread-locality. That is C++ smart_ptr has to be atomic, while its D counter part may safely be non-atomic because everything is TLS be default. Of course, there is a place for smart pointer shared across threads, but I'd personally go with message passing instead. --- Dmitry Olshansky
Jan 14
next sibling parent Martin Nowak <code+news.digitalmars dawg.eu> writes:
On 01/14/2017 04:41 PM, Dmitry Olshansky wrote:
 Not quite true, there is at least one huge advantage - thread-locality.
 That is C++ smart_ptr has to be atomic, while its D counter part may
 safely be non-atomic because everything is TLS be default.
And you can add a shared postblit overload for when it's not thread local ;).
Jan 14
prev sibling parent Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= writes:
On Saturday, 14 January 2017 at 15:41:01 UTC, Dmitry Olshansky 
wrote:
 That is C++ smart_ptr has to be atomic, while its D counter 
 part may safely be non-atomic because everything is TLS be 
 default.
I assume you mean std::shared_ptr. The reference counting semantics are atomic, but the I don't think the compiler is required to if it provably isn't shared. There are also ways to get around it if needed (you only need atomic count when you enter or leave a context, e.g. thread). Of course, there are C++ single threaded alternatives with intrusive ref counting, which I believe is what D is going for. shared_ptr is non-intrusive (doesn't affect allocation or object types).
Jan 23
prev sibling parent reply Chris Wright <dhasenan gmail.com> writes:
On Sat, 14 Jan 2017 02:05:11 +0000, nbro wrote:

 On Friday, 13 January 2017 at 16:50:37 UTC, Dsby wrote:
 I write the ref count pointer and the scoped point in D. it just Like
 cpp's shared_ptr , waek_ptr and unique_ptr .
 Now, it is  Developing.
 I will write more test before the frist release.
 And the docs is null.
 It on github: https://github.com/huntlabs/SmartRef
What's would be the advantages of smart pointers in D?
It's reference counting. Reference counting is like garbage collection, but deamortized. This is better for real-time applications. However, it adds overhead on every assignment and every variable going out of scope. In D, garbage collection is more expensive than it is in other languages, so the tradeoff is more attractive than it would be in other languages.
Jan 13
parent reply nbro <neo matrix.com> writes:
On Saturday, 14 January 2017 at 04:14:11 UTC, Chris Wright wrote:
 It's reference counting.

 Reference counting is like garbage collection, but deamortized. 
 This is better for real-time applications. However, it adds 
 overhead on every assignment and every variable going out of 
 scope.

 In D, garbage collection is more expensive than it is in other 
 languages, so the tradeoff is more attractive than it would be 
 in other languages.
Garbage collection in D is more expensive just because of the poor implementation, from what I've heard. If that's the case, people who work on it should be able to improve it over time.
Jan 14
parent Chris Wright <dhasenan gmail.com> writes:
On Sat, 14 Jan 2017 11:52:34 +0000, nbro wrote:
 Garbage collection in D is more expensive just because of the poor
 implementation, from what I've heard. If that's the case, people who
 work on it should be able to improve it over time.
I posted about this in general. A GC for Java, Python, Ruby, etc can locate type information and GC metadata for an allocation in O(1) time. D's can locate type information and GC metadata in O(log N) time in the worst case, even with the best possible implementation, and its design decisions make the worst case incredibly common. That is one of the common operations that a GC does, so it has a big performance impact. D's GC must be slower because it allows pointers to arbitrary places inside an allocation, and it strongly encourages this with array slicing.
Jan 14
prev sibling parent reply =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
On Friday, 13 January 2017 at 16:50:37 UTC, Dsby wrote:
 I write the ref count pointer and the scoped point in D.
How do two of these differ from - https://dlang.org/phobos/std_typecons.html#.RefCounted - https://dlang.org/phobos/std_typecons.html#.Unique under https://dlang.org/phobos/std_typecons.html ?
Jan 14
parent reply Dsby <dushibaiyu yahoo.com> writes:
On Saturday, 14 January 2017 at 17:35:09 UTC, Nordlöw wrote:
 On Friday, 13 January 2017 at 16:50:37 UTC, Dsby wrote:
 I write the ref count pointer and the scoped point in D.
How do two of these differ from - https://dlang.org/phobos/std_typecons.html#.RefCounted - https://dlang.org/phobos/std_typecons.html#.Unique under https://dlang.org/phobos/std_typecons.html ?
The RefCount not support class or inteface, and the ref count is not atomic , it not thread safe. The Unique is base of GC. It can not nogc. And it use the 'delete' keyword , it will be deprcated(http://dlang.org/deprecate.html). The SmartRef is Base std.experimental.allocator. You can control where the memony allocator. And the smartref.sharedref use the atomic default, you alse can not use atomic。 And have the smartref.weakref with sharedref to fix circular reference.
Jan 15
parent reply Dsby <dushibaiyu yahoo.com> writes:
On Sunday, 15 January 2017 at 15:42:19 UTC, Dsby wrote:
 On Saturday, 14 January 2017 at 17:35:09 UTC, Nordlöw wrote:
 On Friday, 13 January 2017 at 16:50:37 UTC, Dsby wrote:
 I write the ref count pointer and the scoped point in D.
How do two of these differ from - https://dlang.org/phobos/std_typecons.html#.RefCounted - https://dlang.org/phobos/std_typecons.html#.Unique under https://dlang.org/phobos/std_typecons.html ?
The RefCount not support class or inteface, and the ref count is not atomic , it not thread safe. The Unique is base of GC. It can not nogc. And it use the 'delete' keyword , it will be deprcated(http://dlang.org/deprecate.html). The SmartRef is Base std.experimental.allocator. You can control where the memony allocator. And the smartref.sharedref use the atomic default, you alse can not use atomic。 And have the smartref.weakref with sharedref to fix circular reference.
Sorry,In the new phobos, THe Unique is not used the 'delete' keyword. But , it only destroy, not free the memony. and : In https://github.com/dlang/phobos/blob/master/std/typecons.d#L147 ~this() { debug(Unique) writeln("Unique destructor of ", (_p is null)? null: _p); if (_p !is null) destroy(_p); _p = null; } if the 'T' is a struct, it will not exec the Destory function. Is it a bug?
Jan 15
parent reply biozic <dransic gmail.com> writes:
On Sunday, 15 January 2017 at 15:56:30 UTC, Dsby wrote:
 and : In 
 https://github.com/dlang/phobos/blob/master/std/typecons.d#L147
     ~this()
     {
         debug(Unique) writeln("Unique destructor of ", (_p is 
 null)? null: _p);
         if (_p !is null) destroy(_p);
         _p = null;
 }
  if the 'T' is a struct, it will not exec the Destory function. 
 Is it a bug?
What do you mean? This works for me: --- import std.stdio, std.typecons; struct Foo { ~this() { writeln("I'm destroyed"); } } void main() { Unique!Foo foo = new Foo; } // Prints "I'm destroyed" ---
Jan 15
parent reply Dsby <dushibaiyu yahoo.com> writes:
On Sunday, 15 January 2017 at 17:24:25 UTC, biozic wrote:
 On Sunday, 15 January 2017 at 15:56:30 UTC, Dsby wrote:
 and : In 
 https://github.com/dlang/phobos/blob/master/std/typecons.d#L147
     ~this()
     {
         debug(Unique) writeln("Unique destructor of ", (_p is 
 null)? null: _p);
         if (_p !is null) destroy(_p);
         _p = null;
 }
  if the 'T' is a struct, it will not exec the Destory 
 function. Is it a bug?
What do you mean? This works for me: --- import std.stdio, std.typecons; struct Foo { ~this() { writeln("I'm destroyed"); } } void main() { Unique!Foo foo = new Foo; } // Prints "I'm destroyed" ---
the "writeln("I'm destroyed");" not run the ~this in the Unique destroy function. it run in the app exit , THe GC distroy all memony. it example can show : import std.stdio; import std.typecons; struct Foo { ~this() { writeln("I'm destroyed"); } } void fun(){ Unique!Foo foo = new Foo; writeln("exit the fun."); } void main() { fun(); writeln("exit the Main."); } It is the printf: ~/tmp  rdmd ./type.d 2017年01月16日 星期一 09时50分00秒 exit the fun. exit the Main. I'm destroyed ~/tmp  if you use the struct in Unique, the struct's Destory function is not run in the Unique destroy, it is also run in the GC collet. I think it is not the Unique should be.
Jan 15
parent biozic <dransic gmail.com> writes:
On Monday, 16 January 2017 at 01:54:35 UTC, Dsby wrote:
 On Sunday, 15 January 2017 at 17:24:25 UTC, biozic wrote:
 On Sunday, 15 January 2017 at 15:56:30 UTC, Dsby wrote:
 and : In 
 https://github.com/dlang/phobos/blob/master/std/typecons.d#L147
     ~this()
     {
         debug(Unique) writeln("Unique destructor of ", (_p is 
 null)? null: _p);
         if (_p !is null) destroy(_p);
         _p = null;
 }
  if the 'T' is a struct, it will not exec the Destory 
 function. Is it a bug?
What do you mean? This works for me: --- import std.stdio, std.typecons; struct Foo { ~this() { writeln("I'm destroyed"); } } void main() { Unique!Foo foo = new Foo; } // Prints "I'm destroyed" ---
the "writeln("I'm destroyed");" not run the ~this in the Unique destroy function. it run in the app exit , THe GC distroy all memony. it example can show : import std.stdio; import std.typecons; struct Foo { ~this() { writeln("I'm destroyed"); } } void fun(){ Unique!Foo foo = new Foo; writeln("exit the fun."); } void main() { fun(); writeln("exit the Main."); } It is the printf: ~/tmp  rdmd ./type.d 2017年01月16日 星期一 09时50分00秒 exit the fun. exit the Main. I'm destroyed ~/tmp  if you use the struct in Unique, the struct's Destory function is not run in the Unique destroy, it is also run in the GC collet. I think it is not the Unique should be.
Right, good point. This behaviour is indeed caused by destroy() and is not specific to Unique. But it the case of Unique, relying on this (undefined?) behaviour of destroy is a bug (a regression).
Jan 16