Not sure this warrants a breaking change. The whole point of the simple renderers is to provide a fire-and-forget API. For special use cases with options and what not, a custom-built renderer seems to be the way to go.

What use case that’s good for the broader audience does passing the annotation stack have?

Not sure this warrants a breaking change

Fair enough.

On reflection, this is probably more niche than I originally thought. Even in my case, there's a better solution that would involve improving the graphviz API.

Hm, admittedly that's not the resolution that I had expected. It doesn't seem implausible to me that other users of renderSimplyDecorated might want to have access to the annotation stack too. As breaking changes go, I think this one would also have been on the harmless side.

Hm, admittedly that's not the resolution that I had expected.

Yeah, my sudden change of heart mostly comes from the realisation that I probably wouldn't use it anyway in the long run.

As breaking changes go, I think this one would also have been on the harmless side.

Probably true, but it's still a change people would have to adapt to, for no gain in most cases. And prettyprinter has a lot of reverse dependencies, even if the majority of those are probably not using this module since they're not adding backends.

Anyway, this could be reopened, perhaps with the generalised form made in to a separate function, so that it's not a breaking change. And generalised further by e.g. passing the stack to the push and pop functions as well.