Bug 11722 – Qualifier-only casts should not invoke opCast

This is a spin-off issue from bug 5747. D has a cast syntax for modifying just only the type qualifier of the operand. http://dlang.org/expression#CastExpression CastExpression: cast ( Type ) UnaryExpression cast ( CastQual ) UnaryExpression <-- cast ( ) UnaryExpression <-- However, if the operand is class or struct, and has opCast member function, the qualifier-cast cast(qual)exp is translated to cast(qual(typeof(exp))exp, then it will unintendedly invoke opCast. For example, following code wrongly asserts in C.opCast. class C { T opCast(T)() { assert(0); } } void main() { C c = new C(); shared C sc = cast(shared)c; } I think this is not good behavior in generic code, and should be fixed.

Commits pushed to master at https://github.com/D-Programming-Language/dmd https://github.com/D-Programming-Language/dmd/commit/dab021c5b716b3a6eedbb1d883ce55f084b98aca fix Issue 11722 - Qualifier-only casts should not invoke opCast https://github.com/D-Programming-Language/dmd/commit/b32a5205390f69e3fc72b3acb912306f22616fbf Merge pull request #2946 from 9rnsr/fix11722 Issue 11722 - Qualifier-only casts should not invoke opCast

I find the existing behavior reasonable enough, provided it is cast(qual(unqual(typeof(exp)))exp instead of cast(qual(typeof(exp))exp.

I disagree. That was not necessary, unless you prove otherwise.

(In reply to comment #0) > This is a spin-off issue from bug 5747. > > D has a cast syntax for modifying just only the type qualifier of the operand. > > http://dlang.org/expression#CastExpression > > CastExpression: > cast ( Type ) UnaryExpression > cast ( CastQual ) UnaryExpression <-- > cast ( ) UnaryExpression <-- Kenji: I'm wondering: Your original post seemed to imply the enhancement was *only* for qualifier cast. However, the final result seems to be for *any* kind of cast, whith matching types. I mean: const(S) s; auto a = cast() s; auto b = cast(S) s; In the case of "a": We have a "CastQual" type cast, and this bypasses the opCast operator entirelly. In the case of "b": We have a *Type* cast, but that happens to result only in a qualifier change. In the case of "b", *should* we call opCast? Please clarify.

(In reply to comment #5) > (In reply to comment #0) > > This is a spin-off issue from bug 5747. > > > > D has a cast syntax for modifying just only the type qualifier of the operand. > > > > http://dlang.org/expression#CastExpression > > > > CastExpression: > > cast ( Type ) UnaryExpression > > cast ( CastQual ) UnaryExpression <-- > > cast ( ) UnaryExpression <-- > > Kenji: I'm wondering: Your original post seemed to imply the enhancement was > *only* for qualifier cast. However, the final result seems to be for *any* kind > of cast, whith matching types. I mean: > > const(S) s; > auto a = cast() s; > auto b = cast(S) s; > > In the case of "a": We have a "CastQual" type cast, and this bypasses the > opCast operator entirelly. > In the case of "b": We have a *Type* cast, but that happens to result only in a > qualifier change. > > In the case of "b", *should* we call opCast? Please clarify. Yes. And actually, with git-head, following code asserts only in the case of "b". struct S { T opCast(T)() const { assert(0); } } void main() { const(S) s; auto a = cast() s; // bypass opCast auto b = cast(S) s; // call opCast and assert }

(In reply to comment #6) > (In reply to comment #5) > > (In reply to comment #0) > > > This is a spin-off issue from bug 5747. > > > > > > D has a cast syntax for modifying just only the type qualifier of the operand. > > > > > > http://dlang.org/expression#CastExpression > > > > > > CastExpression: > > > cast ( Type ) UnaryExpression > > > cast ( CastQual ) UnaryExpression <-- > > > cast ( ) UnaryExpression <-- > > > > Kenji: I'm wondering: Your original post seemed to imply the enhancement was > > *only* for qualifier cast. However, the final result seems to be for *any* kind > > of cast, whith matching types. I mean: > > > > const(S) s; > > auto a = cast() s; > > auto b = cast(S) s; > > > > In the case of "a": We have a "CastQual" type cast, and this bypasses the > > opCast operator entirelly. > > In the case of "b": We have a *Type* cast, but that happens to result only in a > > qualifier change. > > > > In the case of "b", *should* we call opCast? Please clarify. > > Yes. > And actually, with git-head, following code asserts only in the case of "b". Ah. Oops. I was testing the wrong code. Thanks.