October

We had no regular planning sessions in October. Instead, there were two workgroup meetings focused on DMD as a library. These took place after the monthly meeting and involved several people who were stakeholders, were interested in the project, or had some level of experience with it. The invitation list included Dennis Korpel, Jan Jurzitza (WebFreak), Prajwal S N, Razvan Nitu, Ali Çehreli, Luís Ferreira, Martin Kinkelin, and Walter Bright.

The main focus of these meetings was a proposal Razvan had regarding the possibility of overriding AST Nodes, something he'd brought up in our September meeting. His idea was to enable it via mixins. Walter was reluctant to take that approach and suggested AST nodes should just be simple classes with fields and helper methods. That approach would simplify how AST families are created, but wouldn't fix the underlying issue of being unable to override a node's behavior. (Also see Razvan's DConf '23 talk on the topic.)

In the end, they agreed to refactor the AST implementation by pulling all semantic routines and fields out of the AST nodes so that ASTBase will only contain imports of AST nodes as opposed to the duplicate implementations it contains now. Once this is complete, they'll revisit Razvan's idea about the mixins.

Razvan has since gotten started on this project. You can track its progress in our project tracker. He recently published a blog post as a call to arms seeking assistance. A handful of people have stepped up, but there's still a lot of work. Anyone looking to make some impactful contributions should read the blog post and the relevant section of the contributor guide to learn how to help out.

November

We had one planning session in November. It was held the week before our monthly meeting. The main points of discussion were details about the Bugzilla to GitHub migration and more implementation details for DMD as a library.

For the former, Robert had already migrated the Visual D issues and wanted feedback on the order in which to migrate the remaining projects. Everyone agreed he should go from the project with the lowest issue count to the highest. We also discussed some details about dlang-bot, how to handle attachments, and some more implementation details.

The other issue was regarding extern(C++) classes in DMD. These are extern(C++) so that they're available to LDC, GDC, and any other future tools and compilers that need to integrate the D compiler into a C++ codebase. However, this had caused problems for people using DMD as a library in practice. For example, forgetting that the class is extern(C++) when subclassing it; this caused problems when trying to use features unavailable for extern(C++), e.g., dynamic arrays. There was a workaround for that particular case, but a new one had popped up when the overriding class was templated and was instantiated with, e.g., a dynamic array type.

Razvan expected this would continue to be an annoying problem and was looking for ideas to solve it. We discussed the possibility of making the DMD as a library classes extern(D) instead. Ultimately, the consensus was that though the visitor classes could be made such, there was no simple way to do it for the AST nodes. We have more important issues than this to focus on right now.