Bug 10980 – static initialization of immutable structs with disabled postblit fails

Let S be a struct a disabled potsblit, and we want to create a static instance of an immutable S: //---- alias immutable(S) IS; static IS i = IS.init; //---- 2.060: OK! 2.061: Error: variablei of type struct immutable(S) uses this(this), which is not allowed in static initialization 2.062: Error: variable hello.i of type struct immutable(S) uses this(this), which is not allowed in static initialization 2.063: OK! 2.063.2: OK! So... it broke and it was fixed (apparently). BUT! Let's try that in a template now: //---- template foo(T) { static T i = T.init; alias i foo; } struct S { @disable this(this); } alias immutable(S) IS; //IS i = IS.init; //Uncoment me void main() { auto p = &foo!IS; } //---- 2.060: OK! 2.061: Error: variable main.i of type struct immutable(S) uses this(this), which is not allowed in static initialization 2.062: Error: variable main.i of type struct immutable(S) uses this(this), which is not allowed in static initialization 2.063: Error: template instance main.foo!(immutable(S)) error instantiating 2.063.2: Error: template instance main.foo!(immutable(S)) error instantiating Now, we simply have an "unclear" error that the template failed, but with no diagnostics (at least, I am unable to provide any). Funny story: if we uncomment the global static, it works again. Another workaround is to declare the static inside a function, and have the function return the static by reference. This issue is also "semi-blocker" in emplace, which needs to create a static instance of T, no matter T's type.

(In reply to comment #0) > Let S be a struct a disabled potsblit, and we want to create a static instance > of an immutable S: > > //---- > alias immutable(S) IS; > static IS i = IS.init; > //---- > > 2.060: OK! > 2.061: Error: variablei of type struct immutable(S) uses this(this), which is > not allowed in static initialization > 2.062: Error: variable hello.i of type struct immutable(S) uses this(this), > which is not allowed in static initialization > 2.063: OK! > 2.063.2: OK! > > So... it broke and it was fixed (apparently). The issue had not been fixed, instead it just be hidden by the regression bug 10998, from 2.063. > BUT! > Let's try that in a template now: > > //---- > template foo(T) > { > static T i = T.init; > alias i foo; > } > > struct S > { > @disable this(this); > } > alias immutable(S) IS; > > //IS i = IS.init; //Uncoment me > > void main() > { > auto p = &foo!IS; > } > //---- > 2.060: OK! > 2.061: Error: variable main.i of type struct immutable(S) uses this(this), > which is not allowed in static initialization > 2.062: Error: variable main.i of type struct immutable(S) uses this(this), > which is not allowed in static initialization > 2.063: Error: template instance main.foo!(immutable(S)) error instantiating > 2.063.2: Error: template instance main.foo!(immutable(S)) error instantiating > > Now, we simply have an "unclear" error that the template failed, but with no > diagnostics (at least, I am unable to provide any). > > Funny story: if we uncomment the global static, it works again. Another > workaround is to declare the static inside a function, and have the function > return the static by reference. > > This issue is also "semi-blocker" in emplace, which needs to create a static > instance of T, no matter T's type. Because the current compile-time postblit call check is not enough strict. I opened a compiler fix to improve check mechanism. https://github.com/D-Programming-Language/dmd/pull/2541

Commit pushed to master at https://github.com/D-Programming-Language/dmd https://github.com/D-Programming-Language/dmd/commit/4af52deaba5a24130695a48c132f8e7d46236859 fix Issue 10980 - static initialization of immutable structs with disabled postblit fails