typing: Add a mechanism to disable warnings

[Delja]

Add a mechanism to suppress warnings during the typing phase. I also took the opportunity to modify a bad indentation in the file :smile: .

[Delja]

@Morriar It seems a good idea to put it in the ModelBuilder in order to deactivate all warnings if necessary. The only problem I find is that I have to be careful to set the flag to the correct value each time we run an analysis.

[Delja]

@privat Here is an example of code that the user can provide.

class A
    invariant(bar >= 10)

    var bar: Int
end

class B
    invariant(baz <= 2.0)

    var baz: Float
end

class C
    super A
    super B

    autoinit bar=, baz=
end

In that case, the generated code would look like this

redef class Object
    fun invariant do end
end

class A
    var bar: Int

    redef fun invariant do 
        super
        assert bar >= 10
        print "A"
    end
end

class B
    var baz: Float

    redef fun invariant do 
        super
        assert baz <= 2.0
        print "B"
    end
end

class C
    super A
    super B

    autoinit bar=, baz=
end

var c = new C(11, 1.0)
c.invariant

In that case we get the following error Warning: conflicting property definitions for property _invariant_ in C because class C does not redefine the invariant property.

There are two possibilities in order to solve this problem: introduce a redefinition of _invariant_ which just makes a call to super. Either we authorize the typing without displaying the warnings. I choose not to execute the first solution because it is not necessary to add an additional call when no invariant has been added. I may have been wrong to think it, especially considering the already high cost of invariants

[privat]

Maybe his does not completely solve the issue, but :

There is a is_fictive flag in MEntity that the user should not see (not the best name). See http://nitweb.nitlanguage.org/doc/nitc::MEntity::is_fictive It should be used to tag injected model entities used for implementation shenanigans. This seems the case of contracts. The current effect of fictive entities is that they are not collected (by default) by model filters so they do not appear in docs or metrics. I think that it also make sense that they should not trigger warnings.