Discussion about
http://d.puremagic.com/issues/show_bug.cgi?id=7369
has got me thinking.

On discovering that DMD (2.057) rejects
    invariant() const {}
I had made out that constancy doesn't apply to invariants.  Only later did I realise it does, but requires the syntax
    const invariant() {}

But thinking of it, why should one have to declare invariants as const at all?

Invariants are used to check the integrity of the object, not to modify it, and so it would seem natural that invariants should be automatically const.

The way it is, a type with a non-const invariant cannot have any public const methods.  I can see people sacrificing either invariants or const-correctness to get around this, before they discover the syntax that works for declaring a const invariant.

Notice also that DMD rejects two or more invariants in a single type

    invariant() {}
    const invariant() {}
    immutable invariant() {}

so, unless this is a bug, it isn't in order to have different checks depending on the constancy of either the object or the method called thereon.

What exactly is the intended behaviour here?

And what's your view?  Should we make invariants automatically const?

Stewart.

> On discovering that DMD (2.057) rejects
>     invariant() const {}
> I had made out that constancy doesn't apply to invariants.  Only later did
> I realise it does, but requires the syntax
>     const invariant() {}
Rejecting the first one is just a parser bug.

> Notice also that DMD rejects two or more invariants in a single type
There is an open pull to allow multiple invariants... but not overloading them.  It just concatenates them, like unit tests.

> And what's your view?  Should we make invariants automatically const?
Yes, probably.  Once downside is that any private methods you want to call will need to be const-correct.  Invariants being const is much less restricting than forcing them to be pure/nothrow/safe.

On 29/01/2012 16:46, Daniel Murphy wrote:
<snip>
>> And what's your view?  Should we make invariants automatically const?
> Yes, probably.  Once downside is that any private methods you want to call
> will need to be const-correct.
<snip>

But if you want your class to be const-usable at all, the invariant needs to be const regardless whether we make it automatic or not.

Stewart.

Le 29/01/2012 17:46, Daniel Murphy a écrit :
>> And what's your view?  Should we make invariants automatically const?
> Yes, probably.  Once downside is that any private methods you want to call
> will need to be const-correct.  Invariants being const is much less
> restricting than forcing them to be pure/nothrow/safe.
>

Well, this is required if you want the class to be constable at all, whatever the invariant require it or not. So I'd argue that invraiant should be automatically const.

This is very important to improve D2 and its lib to be more const compliant, and this is a step in the right direction.

On 1/29/2012 10:13 AM, deadalnix wrote:
> This is very important to improve D2 and its lib to be more const compliant, and
> this is a step in the right direction.

I'd have to agree. invariants should always be const.