[SUGGESTION] Aggregate initialization of structs/objects

[realgdman]

I haven't found a way to aggregate initialize struct.

Background: I'm working with Entity-Component-System. It supposes to have many POD structs. Writing constructor in each is cumbersome.
In C++ they could be simply initialized with S s { 1, 7 };
I don't know if I missed correct syntax or it is intended, or is it bug.

S : @struct type = {
public x : int;
public y : float;
}

main: () = {
//s1: S = (1, 2); //cpp2 Ok, cpp1 no matching ctor S s1 {1, 2};

//s2: S;
//s2 = (1, 2); //cpp2 Ok, cpp1 no matching ctor s2.construct(1,2);
}

foo: (inout s3 : S) = {
//s3 = (1, 2); //cpp2 Ok, cpp1 no viable overloaded = s3 = 1, 2;
//s4 = ((1, 2)); //cpp2 Ok, cpp1 no viable overloaded = s4 = 1, 2;
//s5 = (.x=1, .y=2); //I know named not in cpp2 yet
//s6 = {1, 2}; //cpp2 ill formed initializer at {
//also tried for fun s7 = (copy (1, 2) - it emits cpp1 copy(1,2) }`

If it is intended to not have aggregate init, it would be good to have some meta facility, like one autogenerated constructor, which have as parameters all data members in order.

[JohelEGP]

See 1886ed2#r110677033.
You can't declare aggregates in Cpp2 yet.

[realgdman]

Also currently on that assignment, parens got removed, which popped little evil comma operator.

Reproduction example
S : type = {
public x : int = 0;
public y : float = 0;
operator=: (out this, i: int, j: float) = { x = i; y = j; }
operator=: (inout this, i: int) = x = i; //introduced assignment operator
}

main: () = {
s: S = (1, 2);
s = (3, 4); //in cpp1: s = 3, 4; clang warning unused result
std::cout << "(s.x)$; (s.y)$\n"; //print 3; 0.00
}

[JohelEGP]

That's #321.

[realgdman]

Writing constructor in each is cumbersome.

Actually its not allowed to write custom constructor for @struct
error: while applying @struct - a struct may not have a user-defined operator=
So I step back to s: S = (); s.myinit(x, y);

[realgdman]

I have read p0707 proposal, and I see what is meant by @aggregate meta, but I was talking purely about simplicity of use of @struct meta. Because POD structs easy to aggregate initialize in cpp1 by S s { 1, 7 }; compared to my cpp2 s: S = (); s.myinit(1, 7); because I think cpp2 doesnt have that syntax.

[realgdman]

Small proof of concept, only for ints. Before line

cppfront/source/reflect.h

Line 725 in 768f125

basic_value(t); // a plain_struct is-a basic_value

std::string ctor_params = "";
std::string ctor_body = "";
for (auto m : t.get_members()) {
if(m.is_object()) {
ctor_params += ", ctor_" + std::string(m.name()) + ": int";
//TODO reflection +std::string(m.as_object().get_type());
ctor_body += '\n' + std::string(m.name()) + " = ctor_" +
std::string(m.name()) + ";";
}
}
std::string ctor = "operator=: (out this" + ctor_params + ") = {"
+ ctor_body + "}";
t.add_member(ctor);

After which this cpp2 compiles:

S : @struct type = {
x : int = 0;
y : int = 0;
z : i32 = 0;
}

main: () = {
sA: S = (1, 3, 5);
std::cout << sA.y; //prints 3
}

Which generates cpp2
operator=: (out this, ctor_x: int, ctor_y: int, ctor_z: int) = { x = ctor_x; y = ctor_y; z = ctor_z;}

Which emits cpp1
S::S(cpp2::in<int> ctor_x, cpp2::in<int> ctor_y, cpp2::in<int> ctor_z) : x{ ctor_x }, y{ ctor_y }, z{ ctor_z }{}

[realgdman]

Closing in favor of #426 to avoid duplication