Bug 8118 – Impossible to initialize a member struct without default constructor or assigment

struct S { @disable this(); this(int) { } @disable void opAssign(typeof(this)); } class Test { S s = void; // I *EXPLICITLY* told it not to be initialized, but... this() { s = S(to!int("1")); } } void main() { new Test(); } Error: function S.opAssign is not callable because it is annotated with @disable Error: default construction is disabled for type Test Structs without default constructors are pretty much impossible to use.

Possibly related: http://d.puremagic.com/issues/show_bug.cgi?id=8117

(In reply to comment #0) > struct S > { > @disable this(); > this(int) { } > @disable void opAssign(typeof(this)); > } > > class Test > { > S s = void; // I *EXPLICITLY* told it not to be initialized, but... > this() { s = S(to!int("1")); } > } > > void main() { new Test(); } > > > Error: function S.opAssign is not callable because it is annotated with > @disable > Error: default construction is disabled for type Test > > > > Structs without default constructors are pretty much impossible to use. No bug here you just diabled too much. Undisable opAssign. It's opAssign that gets called whne a = ... is seen: this() { s = S(to!int("1")); } If opAssign is trivial it replaced with bitblit. (that is disable comes first!)

(In reply to comment #2) > No bug here you just diabled too much. Undisable opAssign. > > It's opAssign that gets called whne a = ... is seen: > this() { s = S(to!int("1")); } > If opAssign is trivial it replaced with bitblit. (that is disable comes first!) Uh, no, it's a bug IMO. I never asked for an assignment. I want to CONSTRUCT the object manually.

(In reply to comment #3) > (In reply to comment #2) > > No bug here you just diabled too much. Undisable opAssign. > > > > It's opAssign that gets called whne a = ... is seen: > > this() { s = S(to!int("1")); } > > If opAssign is trivial it replaced with bitblit. (that is disable comes first!) > > Uh, no, it's a bug IMO. > > > I never asked for an assignment. I want to CONSTRUCT the object manually. For example, pretend this is the Scoped struct. I *obviously* wouldn't want to assign anything, but I'd want to construct the object. The fact that it's **impossible** to call the constructor directly without an assignment getting in the way (as far as I see) is a bug.

(In reply to comment #4) > The fact that it's **impossible** to call the constructor directly without an > assignment getting in the way (as far as I see) is a bug. I agree with you. I have a struct that is not supposed to be copied. Now I cannot use it as a field in any other struct/class. In some cases a work-around may be to allow assignments, but check that the receiver is S.init. Also I tried "= void" first. So it may be the most intuitive to use for the bug fix. I haven't checked, but it could allow code like this if not currently possible: S s = void; if (xyz) { s = S(3); } else { s = S(7); }

This seems to be fixed now (code compiles and runs on 2.064).

(In reply to comment #6) > This seems to be fixed now (code compiles and runs on 2.064). The OP case is fixed in 2.064, by fixing issue 9665. (In reply to comment #5) > I haven't checked, but it could allow code like this if not currently possible: > > S s = void; > if (xyz) { > s = S(3); > } else { > s = S(7); > } This is completely different case. It would need to use std.conv.emplace. *** This issue has been marked as a duplicate of issue 9665 ***