Bug 17729 – dmd says cast expression is "not an lvalue", but it can be used as one in other contexts

As requested by Andrei on GitHub (<https://github.com/dlang/druntime/pull/1605#discussion_r131770687>). Taking the address of a cast expression gets rejected: ---- struct S { int field = 0; } void main() { shared S s; auto p = & cast(S) s; /* Error: cast(S)s is not an lvalue */ } ---- But we can take the address of the field and it's the same as `s`'s: ---- struct S { int field = 0; } void main() { shared S s; auto p = &(cast(S) s).field; *p = 1; assert(s.field == 1); } ---- We can even use the cast expression on the left-hand side of an assignment: ---- struct S { int field = 0; } void main() { shared S s; cast(S) s = S(1); /* no error */ assert(s.field == 1); /* passes */ } ---- Either all snippets should be rejected, or they should all be accepted.

Furthermore, it doesn't work with ref parameters either: --- struct S { int i; } void fun(ref S s, int i) { s.i = i; } void main() { shared S s; /* This fails: onlineapp.d(17): Error: function onlineapp.fun(ref S s, int i) is not callable using argument types (S, int) onlineapp.d(17): cannot pass rvalue argument cast(S)s of type S to parameter ref S s fun(cast(S) s, 1); */ (*(cast (S*) &s)).fun(1); // Nasty workaround assert (s.i == 1); } --- I also can't understand why the workaround works at all, the result of dereferencing a pointer should surely be a rvalue? Is that also a bug?

(In reply to er.krali from comment #1) > I also can't understand why the workaround works at all, the result of > dereferencing a pointer should surely be a rvalue? > > Is that also a bug? Dereferencing a pointer gives an lvalue. If it gave an rvalue, pointers would be pretty useless. You couldn't assign through them. If you disagree or have more questions on this, I'd suggest making a thread on D.learn [1]. We're going off topic here. [1] https://forum.dlang.org/group/learn

(In reply to ag0aep6g from comment #2) > Dereferencing a pointer gives an lvalue. If it gave an rvalue, pointers > would be pretty useless. You couldn't assign through them. > > If you disagree or have more questions on this, I'd suggest making a thread > on D.learn [1]. We're going off topic here. > > > [1] https://forum.dlang.org/group/learn Sure, now that I think about it, it seems indeed obvious, sorry for the OT.

This has apparently been fixed by the fix for issue 19754. `auto p = & cast(S) s;` compiles now. Closing as duplicate. *** This issue has been marked as a duplicate of issue 19754 ***