www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - Troubles creating templated inout objects

reply Timoses <timosesu gmail.com> writes:
How do I create an inout object with template parameters?

Take following code:

	import std.stdio;
	import std.traits;

	struct S
	{
		int[] arr;
	}

	interface I
	{
		inout(I) opIndex(size_t idx) inout;
	}

	class Test(T) : I
	{
                 // Error: variable 
`onlineapp.Test!(inout(int)[]).Test.member` only parameters or 
stack based variables can be inout
		T member;

		this(inout T mem) inout
		{
			this.member = mem;
		}

		inout(Test!T) get() inout
		{
			return new inout Test!(Unqual!(typeof(member)))(member);
		}

		inout(I) opIndex(size_t idx) inout
		{
			switch (idx)
			static foreach (index, t; T.tupleof)
			{
				case index:
                                         // Error: template 
instance `onlineapp.Test!(inout(int)[])` error instantiating
					return new inout
						Test!(Unqual!(typeof(this.member.tupleof[index])))
									(this.member.tupleof[index]);
                 default:
                 	return null;
			}
		}
	}


	unittest
	{
		auto s = S([1,2,3]);
		auto t = new const Test!S(s);
	}


`Unqual` in this case just turns `inout(int[])` into 
`inout(int)[]`, which is why it complains. That's a side effect 
of this example, however the main question is how one would go 
about achieving something like this idiomatically?

I would like to return a new object and that object should have 
the same mutability as the one creating it.
Jul 10 2018
next sibling parent reply Steven Schveighoffer <schveiguy yahoo.com> writes:
On 7/10/18 10:34 AM, Timoses wrote:
 How do I create an inout object with template parameters?
 
 Take following code:
 
      import std.stdio;
      import std.traits;
 
      struct S
      {
          int[] arr;
      }
 
      interface I
      {
          inout(I) opIndex(size_t idx) inout;
      }
 
      class Test(T) : I
      {
                  // Error: variable 
 `onlineapp.Test!(inout(int)[]).Test.member` only parameters or stack 
 based variables can be inout
          T member;
 
          this(inout T mem) inout
          {
              this.member = mem;
          }
 
          inout(Test!T) get() inout
          {
              return new inout
Test!(Unqual!(typeof(member)))(member);
          }
 
          inout(I) opIndex(size_t idx) inout
          {
              switch (idx)
              static foreach (index, t; T.tupleof)
              {
                  case index:
                               
 Â Â Â Â Â Â Â Â Â  // Error: template instance 
 `onlineapp.Test!(inout(int)[])` error instantiating
                      return new inout
                         
Test!(Unqual!(typeof(this.member.tupleof[index])))
                                     
(this.member.tupleof[index]);
                  default:
                      return null;
              }
          }
      }
 
 
      unittest
      {
          auto s = S([1,2,3]);
          auto t = new const Test!S(s);
      }
 
 
 `Unqual` in this case just turns `inout(int[])` into `inout(int)[]`, 
 which is why it complains. That's a side effect of this example, however 
 the main question is how one would go about achieving something like 
 this idiomatically?
 
 I would like to return a new object and that object should have the same 
 mutability as the one creating it.
You are overthinking :) inout typically is much easier than you expect, until you need to create temporary structs or types with inout members, then it becomes problematic. https://run.dlang.io/is/kosYuC I had to put in a static if, because your function doesn't work once you get down to the array type. See the // fixme comment. -Steve
Jul 10 2018
parent reply Timoses <timosesu gmail.com> writes:
On Tuesday, 10 July 2018 at 18:01:59 UTC, Steven Schveighoffer 
wrote:
 You are overthinking :) inout typically is much easier than you 
 expect, until you need to create temporary structs or types 
 with inout members, then it becomes problematic.

 https://run.dlang.io/is/kosYuC

 I had to put in a static if, because your function doesn't work 
 once you get down to the array type. See the // fixme comment.
Ok, well that helped a tiny bit for the example. I'm trying to reproduce the errors from my project. It's starting to get out of control : D. inout is on a rampage! https://run.dlang.io/is/5TN7XX I guess it's the same as for immutable initialization of arrays. I can't seem to find a proper response to this one.. import std.traits; struct S { int[3] arr; } struct SS { S s; } interface I { inout(I) opIndex(size_t idx) inout; } class Test(T) : I { T member; this(inout T mem) inout { this.member = mem; } inout(Test!T) get() inout { return new inout Test!(Unqual!(typeof(member)))(member); } inout(I) opIndex(size_t idx) inout { static if (is(T == struct)) { switch (idx) static foreach (index, t; T.tupleof) { case index: return new inout Test!(Unqual!(typeof(this.member.tupleof[index]))) (this.member.tupleof[index]); default: return null; } } else return null; } } auto test(T)(inout T t) { return new inout Test!(Unqual!T)(t); } class TestA(T : T[]) { Test!T[] arr; // ERROR: Can't initialize inout variable in a for loop... this(inout(T[]) arr) inout { // 1: Nope foreach (mem; arr) this.arr ~= test(mem); // 2: Nope //Test!T[] a; //foreach (mem; arr) // a ~= test(mem); import std.algorithm : map; // 3: Nope // this.arr = arr.map!((e) => test(e)).array; } } void main() { auto ss = SS(S([1,2,3])); auto t = new const Test!SS(ss); auto ta = new const TestA!(Test!SS[])([t]); }
Jul 11 2018
next sibling parent Timoses <timosesu gmail.com> writes:
On Wednesday, 11 July 2018 at 12:55:35 UTC, Timoses wrote:
 On Tuesday, 10 July 2018 at 18:01:59 UTC, Steven Schveighoffer 
 wrote:
 You are overthinking :) inout typically is much easier than 
 you expect, until you need to create temporary structs or 
 types with inout members, then it becomes problematic.

 https://run.dlang.io/is/kosYuC

 I had to put in a static if, because your function doesn't 
 work once you get down to the array type. See the // fixme 
 comment.
Ok, well that helped a tiny bit for the example. I'm trying to reproduce the errors from my project. It's starting to get out of control : D. inout is on a rampage! https://run.dlang.io/is/5TN7XX I guess it's the same as for immutable initialization of arrays. I can't seem to find a proper response to this one.. [...] class TestA(T : T[]) { Test!T[] arr; // ERROR: Can't initialize inout variable in a for loop... this(inout(T[]) arr) inout { // 1: Nope foreach (mem; arr) this.arr ~= test(mem); // 2: Nope //Test!T[] a; //foreach (mem; arr) // a ~= test(mem); import std.algorithm : map; // 3: Nope // this.arr = arr.map!((e) => test(e)).array; } } [...]
I guess the problem here is focused around the problem that the incoming type in the constructor is inout and that the constructed object itself is inout. I can't seem to find another way, I'm just blatantly casting now... class TestA(T : T[]) { Test!T[] arr; this(inout(T[]) arr) inout { import std.algorithm : map; import std.range: array; // should be okay to cast to const, won't change anything auto ts = cast(const T[])arr; // should be okay as well, as `test(t)` creates a new object this.arr = cast(inout(Test!T[]))(ts.map!(t => test(t)).array); } } I also found `assumeUnique` in std.exception, or std.experimental.allocator.makeArray. I'm not to sure how they might be able to circumvent the casting though, since I need an inout array of the objects... Am I missing something or is `inout` simply not that well "implemented" yet?
Jul 12 2018
prev sibling parent reply Steven Schveighoffer <schveiguy yahoo.com> writes:
On 7/11/18 8:55 AM, Timoses wrote:
 On Tuesday, 10 July 2018 at 18:01:59 UTC, Steven Schveighoffer wrote:
 You are overthinking :) inout typically is much easier than you 
 expect, until you need to create temporary structs or types with inout 
 members, then it becomes problematic.

 https://run.dlang.io/is/kosYuC

 I had to put in a static if, because your function doesn't work once 
 you get down to the array type. See the // fixme comment.
Ok, well that helped a tiny bit for the example. I'm trying to reproduce the errors from my project. It's starting to get out of control : D. inout is on a rampage! https://run.dlang.io/is/5TN7XX I guess it's the same as for immutable initialization of arrays. I can't seem to find a proper response to this one..     import std.traits;     struct S     {         int[3] arr;     }     struct SS     {         S s;     }     interface I     {         inout(I) opIndex(size_t idx) inout;     }     class Test(T) : I     {         T member;         this(inout T mem) inout         {             this.member = mem;         }         inout(Test!T) get() inout         {             return new inout Test!(Unqual!(typeof(member)))(member);         }         inout(I) opIndex(size_t idx) inout         {             static if (is(T == struct))             {                 switch (idx)                 static foreach (index, t; T.tupleof)                 {                     case index:                         return new inout Test!(Unqual!(typeof(this.member.tupleof[index])))                                Â Â Â Â Â Â Â Â Â  (this.member.tupleof[index]);                     default:                         return null;                 }             }             else                 return null;         }     }     auto test(T)(inout T t)     {         return new inout Test!(Unqual!T)(t);     }     class TestA(T : T[])     {         Test!T[] arr;                 // ERROR: Can't initialize inout variable in a for loop...         this(inout(T[]) arr) inout         {             // 1: Nope             foreach (mem; arr)                 this.arr ~= test(mem);             // 2: Nope             //Test!T[] a;             //foreach (mem; arr)             //   a ~= test(mem);
On the right track, but inside inout (or const or immutable) constructors, the members can only be initialized once. So you have to initialize a local, and then set the member once. The issue is, your input is *also* inout (a necessary condition), so you didn't declare a properly: inout(Test!T)[] a; foreach (mem; arr) a ~= test(mem); this.arr = a; -Steve
Jul 12 2018
parent Timoses <timosesu gmail.com> writes:
On Thursday, 12 July 2018 at 12:22:34 UTC, Steven Schveighoffer 
wrote:
 On 7/11/18 8:55 AM, Timoses wrote:
      class TestA(T : T[])
      {
          Test!T[] arr;
 
                  // ERROR: Can't initialize inout variable in 
 a for loop...
          this(inout(T[]) arr) inout
          {
              // 1: Nope
              foreach (mem; arr)
                  this.arr ~= test(mem);
 
              // 2: Nope
              //Test!T[] a;
              //foreach (mem; arr)
              //   a ~= test(mem);
On the right track, but inside inout (or const or immutable) constructors, the members can only be initialized once. So you have to initialize a local, and then set the member once. The issue is, your input is *also* inout (a necessary condition), so you didn't declare a properly: inout(Test!T)[] a; foreach (mem; arr) a ~= test(mem); this.arr = a; -Steve
Aw, thanks! This is much nicer than casting...
Jul 12 2018
prev sibling parent reply Timoses <timosesu gmail.com> writes:
On Tuesday, 10 July 2018 at 14:34:55 UTC, Timoses wrote:
 `Unqual` in this case just turns `inout(int[])` into 
 `inout(int)[]`, which is why it complains. That's a side effect 
 of this example [...]
See also: https://issues.dlang.org/show_bug.cgi?id=3567
Jul 12 2018
parent Steven Schveighoffer <schveiguy yahoo.com> writes:
On 7/12/18 4:58 AM, Timoses wrote:
 On Tuesday, 10 July 2018 at 14:34:55 UTC, Timoses wrote:
 `Unqual` in this case just turns `inout(int[])` into `inout(int)[]`, 
 which is why it complains. That's a side effect of this example [...]
See also: https://issues.dlang.org/show_bug.cgi?id=3567
Responded there, but really this is a misunderstanding of Unqual. Unqual should shallowly strip as much off as it can, but shouldn't allow you to break const. The problem is that for things other than pointers or arrays, it can't provide a tail-const version, so it goes too far. Fixing this will probably break a lot of code, which means we need to find another way. -Steve
Jul 12 2018