Bug 10993 – mangling of voldemort types with lambdas changes during return type inference

Reduced from the cartesianProduct unittests: module test3; //import std.traits; auto foo(T)(T a) { static immutable typeof(a) q; pragma(msg, "foo: " ~ typeof(q).mangleof); return q; } struct test(alias fn) { bool ini = true; void* p; } auto fun() { auto x = foo!()(test!(a=>a)()); pragma(msg, "fun: " ~ typeof(x).mangleof); return x; //typeid(x); } void main() { const x = fun(); pragma(msg, "x : " ~ typeof(x).mangleof); auto y = cast()x; pragma(msg, "y : " ~ typeof(y).mangleof); } compile with "dmd test3.d" yields: foo: yS5test33fun32__T4testS205test33fun9__lambda2Z4test fun: yS5test33fun32__T4testS205test33fun9__lambda2Z4test x : yS5test33fun32__T4testS205test33fun9__lambda2Z4test y : S5test33funFZyS5test33fun32__T4testS205test33fun9__lambda2Z4test32__T4testS 205test33fun9__lambda2Z4test i.e. casting away immutable leads to a completely new mangling while it should have removed the 'y' at the start only. This causes link errors when trying to produce complete TypeInfo data.

Commit pushed to master at https://github.com/D-Programming-Language/dmd https://github.com/D-Programming-Language/dmd/commit/be57510ce308da696ed293d4acc4b233f3258b30 Merge pull request #2585 from rainers/issue10993 fix Issue 10993: mangling of voldemort types with lambdas changes during return type inference