Bug 15936 – misleading error message about @disable where there is none

Found by denizzzka who posted to D.learn: http://forum.dlang.org/post/[email protected] ---- struct Vector { void opAssign(Vector u) {} } struct RAMNode { Vector min; const bool isLeafNode = false; } void main() { RAMNode old_root; old_root = RAMNode.init; /* line 15 */ } ---- This code can't compile because isLeafNode is const. dmd gets confused by the opAssign, though, and prints: "test.d(15): Error: function test.RAMNode.opAssign is not callable because it is annotated with @disable". That's nonsense. There is no @disable anywhere. Without the opAssign the error message is ok: "test.d(15): Error: cannot modify struct old_root RAMNode with immutable members".

Yeah that's a compiler implementation detail. DMD tries to instantiate the code with gagging on, and on error, it just marks it as '@disable'd. For reference, the guilty code is https://github.com/dlang/dmd/blob/87ce5f860b9b8c4a9280e9d9d8643ac6a0df7bbd/src/clone.d#L319-L324

The error message has improved: test.d(15): Error: generated function `test.RAMNode.opAssign` cannot be used because it is annotated with `@disable` Since the error happens in generated code, the compiler does not output it. This makes sense as the user cannot directly modify generated code. So it just informs that the generated code has failed. Ideally, the compiler would point out an actionable item for the user to fix the breakage, however, currently the rudimentary error handling in dmd does not allow for that. Closing as WORKSFORME.