Bug 10413 – .init incorrectly accepts any expression

According to the spec, only variables, fields, and types support the .init property. But all kinds of other things compile: int foo() { return 1; } static assert(foo.init == 0); This gets changed into typeof(foo()).init. --- diff --git a/src/mtype.c b/src/mtype.c index 1c83eb9..8217e8d 100644 --- a/src/mtype.c +++ b/src/mtype.c @@ -2051,6 +2051,11 @@ Expression *Type::dotExp(Scope *sc, Expression *e, Identifier *ident, int flag) char *s = e->toChars(); e = new StringExp(e->loc, s, strlen(s), 'c'); } + else if (ident == Id::init && ex->op != TOKtype) + { + error(e->loc, ".init can only be applied to variables, fields, and types. Did you mean typeof(%s).init ?", e->toChars()); + e = new ErrorExp(); + } else e = getProperty(e->loc, ident, flag);

Pull request: https://github.com/D-Programming-Language/dmd/pull/2240 This fails Phobos unit tests in multiple places, for example: std/traits.d(3654): Error: static assert (is(void == double)) is false With constructions like: template Foo(T...) T[0].init is OK if T[0] is a type, but if it is a value, it won't compile. It is in fact pretty bizarre that 2.init == 0, so this should probably be changed.

(In reply to comment #1) > With constructions like: > template Foo(T...) > > T[0].init > is OK if T[0] is a type, but if it is a value, it won't compile. If the template code doesn't make sense, then it doesn't compile. Seems fine to me. > It is in fact pretty bizarre that 2.init == 0, so this should probably be > changed. ".init" is a static property, and as such, can be called on either type or instance, just the same way you can write "assert(1.max == 0x7FFFFFF)". As a matter of fact, I'm pretty sure I've seen this used before, instance of having to pulling out the "typeof". I'm not sure changing this would be correct.

(In reply to comment #2) > instance of having > to pulling out the "typeof". I mean "instead of having to pull out the "typeof"."

According to Walter it's ok if a UDT defines its own .init, which would make value.init valid: https://issues.dlang.org/show_bug.cgi?id=7066 I don't agree with him though, especially since the whole argument against allowing user-defined default ctors is the compile-time known .init property.

int foo() { return 1; } static assert(foo.init == 0); is a case of optional parentheses, as char foo() { return 1; } static assert(foo.init == 255); passes, but int foo(int) { return 1; } static assert(foo.init == 0); fails with onlineapp.d(3): Error: function onlineapp.foo(int) is not callable using argument types () onlineapp.d(3): missing argument for parameter #1: int onlineapp.d(3): while evaluating: static assert((__error).init == 0) closing as invalid