On Tuesday, 14 November 2023 at 08:18:20 UTC, Mike Parker wrote:
> In September 2023, we had one planning session. The major item on the agenda was editions. Other items were a new meeting format, the Bugzilla to GitHub migration, and the future of D.
Attendees
The following people attended the session.
- Walter Bright
- Ali Çehreli
- Martin Kinkelin
- Dennis Korpel
- Átila Neves
- Razvan Nitu
- Mike Parker
- Robert Schadek
Editions
We had agreed in the September monthly meeting the week before that we need to define what editions will look like before we start deciding which features should go in any given edition. My goal with this session was to establish the first point on the timeline: a deadline for the editions proposal. I also thought it would be a good opportunity for all of us to clarify what editions are (there were some different ideas about that) and discuss aspects of the concept we need to consider.
These are excellent progress. I don't wholly agree with the strong commitment on backwards compability DLF decided to make, but with editions it doesn't matter much. The langauge can remain both backwards compatible and develop further at the same time.
> Here are some points that came out of the discussion.
- Editions are essentially feature sets. Each edition can add/remove/deprecate features.
- Editions are entirely opt-in and only affect the source you explicitly apply them to, i.e., they are not transitive to dependencies.
Makes sense, but on the other hand we will have a lot of questions about templates that are defined in domain of one language edition and instantiated in another. It's going to be onerous to figure out how those cases will work. Regardless, I'm looking forward to this.
>
- Editions will most likely be implemented via an attribute on the module declaration. We haven't discussed any details about that, but for now, just imagine something like
@edition(2024) module foo;.
The syntax isn't critical anyway. That is one small detail that won't have to dominate the discussions.
>
- Features cannot be opted into individually. When you apply an edition to a module, you get the whole thing.
Makes sense again. Lifts a lot of maintenance burden for relatively little user inconvenience. OTOH maybe the existing preview/revert flags should stay for the starting version.
>
- The default edition, meaning the code you have now, should compile forever.
- We should have a tool that automates as much as possible the migration of modules to new editions
Isn't this what the -transition switches do? Regardless, mostly agreed.
About the question of what would be the default edition. Like others I tend to think that in cases where neither the module nor the compiler invocation has any definitions it should be the latest stable edition. Why? Adding -edition=2023 (or whatever the flag for the starting edition would be) to the build script of a legacy project is dead simple, and besides needs to be done only once, after which the project will always compile. People don't hate breakage that much.
The reverse situation, that we'd have to always either use -edition=20xx or explicilty declare the edition in the module header for new code would have the same problems that our @system impure defaults for function attributes. At best, it'd annoy people but more likely it'd more or less lead to new code being written in the starting edition because we don't remember or bother with picking the edition.
Permalink
Reply
