On 10/7/23 20:49, IchorDev wrote:
> I haven’t seen anyone mention this yet…
> 
> https://github.com/dlang/DIPs/pull/234#issuecomment-1751054587
> 
> This is exciting, but also a bit worrying.
> 
> I’m never going to be able to reasonably support multiple editions of the language at once, so I’d likely always support the newest one. I’m worried that this change might fracture D’s library developers.

The intention is the opposite. The vision is actually that library authors only have to support one edition of the language at a time. Code written for different editions can interoperate in the same compiler invocation as they will be based on a compatible internal AST representation.

So basically, everyone updates (or does not update) their editions at their own schedule and old code keeps working. In particular, libraries written for older editions can be seamlessly used with new code written in the latest and shiniest edition. It is a bit like feature flags, but they are enabled per source file instead of per compiler invocation. Another analogy is that it is like ImportC, but instead of importing C code, you import code from another D edition.

This allows editions to freely introduce breaking language changes without breaking the ecosystem at all.

There will be much less pressure on library authors to keep their libraries up to date with the latest language changes.

In practice, of course I anticipate there being some leftover friction, for example, there will certainly be some bugs (or regressions due to bug fixes) and limitations that only affect some combinations of editions, but overall I think both library authors and compiler writers/language designers will be in a much better place than before.