I think we can do this gradually by:

• Subclassing PhabricatorApplicationTransactionEditor into ...Pro or ...ModularEditor or maybe just calling the new one PhabricatorTransactionEditor or even PhabricatorEditor given how core the abstraction is to modern applications. If we do a goofy one, we should rename at the end.
• Subclassing PhabricatorTransaction in the same way.
• Implement all behaviors in terms of modular transaction subclasses.
• Migrate an application.
• Migrate everything else.
• Throw the old stuff away.

Every method should already have all the parameters it needs and all the required context, I think. This isn't really shifting any responsibilities, just reorganizing code.

I'm continuing to warm to this idea and am considering doing it in the relatively near term, either as part of T9132 or directly afterward. In particular, that will also require touching a bunch of applications in a big, related modernization effort, so it may be worthwhile to double up on testing and do this too.

On the other hand, the modernization implied by T9132 is likely much smaller (and much higher-value in the short term per effort spent) than this one.

I'm leaning toward exploring this in the near term. We have four applications which may convert soon:

• Calendar is prioritized and certainly converting. (T9275)
• Differential is probably converting? (T11114)
• Dashboards are probably converting? (T10855)
• Packages may be coming into existence? (T5055)

Additionally, T11035 is prioritized and would be cleaner alongside structural changes.

I'd feel much better about all of this if transactions were modularized first.

One concern here is that Differential, Dashboards and Calendar are all relatively poor starter applications (all have complex, unique transaction behaviors) and Packages doesn't exist yet, and building purely as a testbed for modular transactions feels questionable. So maybe I'll just convert Paste (the canonical starter application) and then go from there. Some weak adjacency to T7643.

I spent a couple hours on this last night and a couple hours this morning and I think I have Paste pretty much converted, so things look promising. I'm going to hold this until after the release cut, but expect to put Calendar on top of it and whatever else converts in the future.

(Just bookkeeping; there's no external priority on this.)

I think we broke plain-text emails here: