Me trying to explain the problem behind #ReflectiveAccessToNonExportedTypes

[Stephane Epardaud]

On 20/07/16 16:33, mark.reinhold at oracle.com wrote:
> Requiring something to be declared explicitly one way or the other, 
> rather than offer a safe and sensible default, seems very unlike the 
> rest of the language. 

I have to say I agree with that, even though I also understand the 
problem that non-public packages will cause to adopters (which is why I 
proposed a meta-annotation solution to _some_ cases).

The default visibility should indeed be (module-)private, that's what it 
is for regular Java declarations, and that's what it should be for 
packages now that they can have visibility.

Even if we required that package declarations (in package-info.java) now 
be declared with `public` modifiers for that (rather than in the module 
descriptor), it would _still_ cause existing packages to become 
non-exported when you convert your jar into a module.

If you don't convert your jar to a module all your packages remain 
exported. If you do, you have to do _something_ to your packages 
whatever solution you pick.

You can say that they should be exported by default, so that existing 
applications using reflection keep working, but that's really a bad 
default. Sometimes backwards-compatibility requires compromises and 
introducing bad defaults, but in this case you already have to do the 
manual step of adding a module descriptor, so at this point it is not 
unreasonable to have the good default of non-exported modules and teach 
people that when they add a module descriptor they also have to decide 
and explicitly state which packages to export.

Now, granted I'm already used to that in both Ceylon and OSGi, so 
perhaps I'm biased.