fix issue 10770 - is(T BASE==enum) with tag enum T - AssertFail:'type' line 428 declaration.c

[hpohl]

http://d.puremagic.com/issues/show_bug.cgi?id=10770

[ghost]

There is an edge case, the following currently works but will fail with your pull:

enum E : int;
static assert(is(E e == enum));

[ghost]

Using the following works for me:

tded = ((TypeEnum *)targ)->toBasetype();

With these test-cases:

enum E : int;
static assert(is(E e == enum));

enum F;
static assert(is(F e == enum));

Edit: Actually they both pass now, but you may be right that the second case should actually be disallowed.

[ghost]

@9rnsr: What do you think, should both cases be allowed? enum F's basetype will be int in the second static assert.

[9rnsr]

In the enum F;, the base type is unknown. Therefore, I think the first case should be allowed, but the second case should not.

[hpohl]

Updated.

[9rnsr]

Adding hasexplicittype looks weird to me.

I had thought that, if !members && !memtype, the taking the base type would cause "forward reference" or "unknown size" error.
However by looking enum.c now, I noticed that EnumDeclaration::getMemtype returns Type::tint32 for the base type of enum E;.

I think that the current EnumDeclaration::getMemtype behavior would be a bug. Could you consider fixing the function rather than adding hasexplicittype field?

[9rnsr]

Sorry for late reply.

While testing your PR in my local, I noticed that it would be better not to cause "forward reference" error in TypeEnum::toBasetype() if the enum type is opaque. So it would become:

 Type *TypeEnum::toBasetype()
 {
+    if (!sym->members && !sym->memtype)
+        return this;
     return sym->getMemtype(Loc())->toBasetype();
 }

[9rnsr]

Please fix file name: test/fail_compilation/10770.d -> test/fail_compilation/ice10770.d

[9rnsr]

@hpohl , Could you please add some comment when you update/push commits? If not, I cannot notice your change (Github doesn't send any notification emails to the review participants in such cases).

[9rnsr]

Do you have any plan to fix issue 10770? If not, I'd open a new PR for the ICE bug.

[9rnsr]

Posted #2795

[9rnsr]

#2795 was merged, so I close this.

[hpohl]

Sorry, will look into my other PRs next weekend.