Thread overview
record: C# like records for D
Jul 14, 2021
Dylan Graham
Jul 15, 2021
Dylan Graham
Jul 15, 2021
Dylan Graham
Jul 16, 2021
Dylan Graham
Jul 16, 2021
Dylan Graham
Jul 16, 2021
vit
Jul 16, 2021
Dylan Graham
Jul 16, 2021
Dylan Graham
July 14, 2021

DUB
Github

This is record. It aims to implement records similar to what C# has by leveraging D's metaprogramming. C# Example 1 C# Example 2.

Future steps are going to be adding a record struct; default value support; init-only-setters like in C#, wherein at the end of construction or duplication, the init lambda for the field is called, and the field can be set that once.

Example:

import drecord;

alias MyRecord = record!(
    get!(int, "x"), /// x is an int, can only be set during construction
    get_set!(float, "y"), /// y is a float, can be get or set whenever
    property!("getDoubleOfX", (r) => r.x * 2), /// a property that returns the double of x
    property!("getMultipleOfX", (r, m) => r.x * m, int), /// that takes an argument and multiples x by that value
    property!("printY", (r) => writeln(r.y)), /// prints y
    property!("resetY", (r) => r.y = 0) /// resets y to 0f
);

auto r = new MyRecord(12, 4.5f); /// sets x, y

writeln(r); // { x = 12, y = 4.5f }
writeln(r.toHash); // 376

writeln(r.x); // 12
writeln(r.getDoubleOfX); // 24
writeln(r.getMultipleOfX(4)); // 48
r.printY; // 4.5
r.resetY;
writeln(r.y); // 0
r.y = 13f;
r.printY; // 13

/// Duplicate r, and set x to 17 (we can only do this in ctor, or during duplication)
/// This is equivalent to C#'s "with" syntax for records
auto q = r.duplicate!("x")(17);
writeln(q); // {x = 17, y = 0}
writeln(q == r); // false
writeln(q is r); // false

auto b = r.duplicate; // duplicate, don't change any fields
writeln(b == r); // true
writeln(b is r); // false
July 15, 2021

On Wednesday, 14 July 2021 at 23:16:05 UTC, Dylan Graham wrote:

>

DUB
Github

record now has support for custom default initialisers. Example:

import drecord;

alias DefaultRecord = record!(
    // The third parameter is a lambda which provides default initialisation
    get!(int, "x", () => 4), // x is set to 4 by default
    get_set(Object, "o", () => new Object) // o is set to a new Object by default
);

auto r = new DefaultRecord; // run the default initialisers
writeln(r); // {x = 4, o = object.Object}

auto q = DefaultRecord.create!"x"(9); // run default initialisers, then set x to 9
writeln(r); // {x = 9, o = object.Object}
July 15, 2021

On Wednesday, 14 July 2021 at 23:16:05 UTC, Dylan Graham wrote:

>

... init-only-setters like in C#, wherein at the end of construction or duplication, the init lambda for the field is called, and the field can be set that once.

This has been implemented with get_compute. Example:

alias MyRecord = record!(
    get!(int, "x", () => 20),
    // get_compute lets you compute a field after the rest have been initialised
    get_compute!(float, "y", (rec) => rec.x * 2f)
);

auto r = new MyRecord;
writeln(r); // {x = 20, y = 40f}
r = new MyRecord(10);
writeln(r); // {x = 10, y = 20f}
r = MyRecord.create!"x"(5);
writeln(r); // {x = 5, y = 10f}
auto q = r.duplicate!"x"(2);
writeln(q); // {x = 2, y = 4f}
July 16, 2021

On Wednesday, 14 July 2021 at 23:16:05 UTC, Dylan Graham wrote:

>

DUB
Github

Found and squashed some critical bugs. Thanks to Adam and Rikki for the help.

Before, record would throw a compilation error due when passed types declared outside of drecord or its imports. Example:

module myapp;

class A{}
auto MyRecord = record!(get!(A, "a")); // would throw an error as it could not find A

This was due to improper usage of .stringof and mixins. This has been corrected and replaced with idiomatic D. The string generation functions for the constructor and opEquals have been removed and replaced in accordance to above.

July 16, 2021

On Friday, 16 July 2021 at 13:14:22 UTC, Dylan Graham wrote:

>

On Wednesday, 14 July 2021 at 23:16:05 UTC, Dylan Graham wrote:

>

DUB
Github

module myapp;

class A{}
auto MyRecord = record!(get!(A, "a")); // would throw an error as it could not find A

That should read

[...]
alias MyRecord = record!(get!(A, "a")); // would throw an error
July 16, 2021

On Wednesday, 14 July 2021 at 23:16:05 UTC, Dylan Graham wrote:

>

DUB
Github

This is record. It aims to implement records similar to what C# has by leveraging D's metaprogramming. C# Example 1 C# Example 2.

Future steps are going to be adding a record struct; default value support; init-only-setters like in C#, wherein at the end of construction or duplication, the init lambda for the field is called, and the field can be set that once.

What adventage has record over normal immutable/const class?

July 16, 2021

On Friday, 16 July 2021 at 13:54:36 UTC, vit wrote:

>

On Wednesday, 14 July 2021 at 23:16:05 UTC, Dylan Graham wrote:

>

DUB
Github

This is record. It aims to implement records similar to what C# has by leveraging D's metaprogramming. C# Example 1 C# Example 2.

Future steps are going to be adding a record struct; default value support; init-only-setters like in C#, wherein at the end of construction or duplication, the init lambda for the field is called, and the field can be set that once.

What adventage has record over normal immutable/const class?

In terms of mutability, none.

The duplicate method, however, lets you copy and mutate (once at duplication) a record without impacting the integrity of the original. You can do this with an immutable class, but then you have to roll your own.

Really, it's more a convenience / less typing sort of feature, as it implements all the common boilerplate for you, which should help with consistency of code.

July 16, 2021

On 7/16/21 10:52 AM, Dylan Graham wrote:

>

On Friday, 16 July 2021 at 13:54:36 UTC, vit wrote:

>

What adventage has record over normal immutable/const class?

In terms of mutability, none.

The duplicate method, however, lets you copy and mutate (once at duplication) a record without impacting the integrity of the original. You can do this with an immutable class, but then you have to roll your own.

Really, it's more a convenience / less typing sort of feature, as it implements all the common boilerplate for you, which should help with consistency of code.

What about a UFCS duplicate method?

I'm not doubting there are good reasons to define types this way in C#, but D has pretty good tools when it comes to boilerplate.

I would possibly suggest that instead of a record template that accepts directives using inline lambdas, etc, just accept a model type and use udas to adjust the record type.

i.e.:

struct RecModel {
   @get_set float y;
   @get int x;
   auto getDoubleOfX() { return x * 2; }
   ... // etc
}

alias MyRecord = record!RecModel;

I use this kind of thing to great success in my SQL database system.

-Steve

July 16, 2021

On Friday, 16 July 2021 at 19:37:53 UTC, Steven Schveighoffer wrote:

>

On 7/16/21 10:52 AM, Dylan Graham wrote:

>

On Friday, 16 July 2021 at 13:54:36 UTC, vit wrote:

>

What adventage has record over normal immutable/const class?

In terms of mutability, none.

The duplicate method, however, lets you copy and mutate (once at duplication) a record without impacting the integrity of the original. You can do this with an immutable class, but then you have to roll your own.

Really, it's more a convenience / less typing sort of feature, as it implements all the common boilerplate for you, which should help with consistency of code.

What about a UFCS duplicate method?

I'm not doubting there are good reasons to define types this way in C#, but D has pretty good tools when it comes to boilerplate.

I would possibly suggest that instead of a record template that accepts directives using inline lambdas, etc, just accept a model type and use udas to adjust the record type.

i.e.:

struct RecModel {
   @get_set float y;
   @get int x;
   auto getDoubleOfX() { return x * 2; }
   ... // etc
}

alias MyRecord = record!RecModel;

I use this kind of thing to great success in my SQL database system.

-Steve

That is a good idea, and to be honest I haven't looked at it that way. So the record is a separate type from its model? Ie: class MyRecord {} based off struct RecordModel {}? Or did I misunderstand?

With regards to things like properties and methods, do you have RecModel inlined in the class and then forward the calls to it? Ie:

struct RecModel {
    @get int x;
    auto doubleOfX() { return x; }
}

class Rec {
    private RecModel instance;

    @property auto x() { return instance.x; }
    auto doubleOfX() { return instance.doubleOfX; }
}

I do like the lambda directives as it, in my mind at least, enforces the idea that the record/class must be a simple data type (ie no crazy methods and such).

I do have to admit this was more an exercise in metaprogramming, and since I did manage to make something I figured I'd share it.

July 16, 2021

On 7/16/21 4:11 PM, Dylan Graham wrote:

>

On Friday, 16 July 2021 at 19:37:53 UTC, Steven Schveighoffer wrote:

>

I would possibly suggest that instead of a record template that accepts directives using inline lambdas, etc, just accept a model type and use udas to adjust the record type.

>

That is a good idea, and to be honest I haven't looked at it that way. So the record is a separate type from its model? Ie: class MyRecord {} based off struct RecordModel {}? Or did I misunderstand?

With regards to things like properties and methods, do you have RecModel inlined in the class and then forward the calls to it? Ie:

struct RecModel {
     @get int x;
     auto doubleOfX() { return x; }
}

class Rec {
     private RecModel instance;

     @property auto x() { return instance.x; }
     auto doubleOfX() { return instance.doubleOfX; }
}

Yeah, something like that. Though there are multiple ways to go about it. One is to write a string that represents what you want it to do, and then mixin that thing. The way I do it is using opDispatch which is super-powerful for forwarding calls like that. I'm assuming you are already doing something like this anyway! It's just, what are the instructions?

I really can't wait to reveal the sql library I've been nursing along with my web project. I don't want to post it yet here, because it will just garner more questions.

I can point you at some early ideas I had on this kind of model-based programming, in a Boston meetup I did a talk for a number of years ago: https://www.youtube.com/watch?v=ZxzczSDaobw

>

I do like the lambda directives as it, in my mind at least, enforces the idea that the record/class must be a simple data type (ie no crazy methods and such).

I'm not sure what you think is a "crazy" method, but lambdas can do whatever a method can do.

I like using models vs. template directives because you get to use the actual language to enforce your "model" is sane, and to make readable instructions for the metaprogramming processor. Like in one of your methods, you have:

property!("getMultipleOfX", (r, m) => r.x * m, int)

What is that int for? It's not clear from the usage, I have to go look it up. Most likely, it's the return type, but it's not nearly as clear as:

int getMultipleOfX(int m) { return x * m; }

Plus, I can debug my model without having to debug the metaprogramming stuff.

>

I do have to admit this was more an exercise in metaprogramming, and since I did manage to make something I figured I'd share it.

It's cool, I love using metaprogramming to write code for me. It's without a doubt the best reason to use D. Learning how to use to avoid writing boilerplate is life changing!

-Steve