You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Design] Discussing the API Expose for the Pluggable Optimization Pass #465
Now that since we already have the pluggable optimization pass, the next step is to discuss what is the intended "proper way" to use it.
Things I can think of:
For different usages, where should the developers place their code?
I can think of 2 example usages of this feature:
ASAPPlanner developers, in optimization algorithm design, may want to maintain some other optimization passes besides the major one within the repo for a long-term development. This requires some places in the repo to hold such separate passes.
ASAPPlanner users may only want to leverage the ASAPPlanner's DAG language, but implement their own optimization passes. For such users, they should implement their optimization outside of the planner repo. Although currently the DAG language themselves are public to users, some useful helper functions, like lift and relational_join_guarantee in ASAPPlanner/crates/asap-aware-mapping/src/replacement.rs, are still private, which is not ideal.
API publicity related to this issue
According to item 2 proposed in the last section, ideally we should expose some helper functions to more "gray-box" lib users.
Should the API publicity for these users be of the same level of the black-box users (who only want to use the end-to-end optimization function)
I see this being useful for ASAPPlanner but we should only do this as necessary. If its not a big concern right now, I would not do anything about it right now, since pre-mature abstraction is more likely to hurt than help.
Activity
milindsrivastava1997 commentedon Sep 25, 2026
@Selvomega We can have a directory in the repo that holds these different optimization passes. That's not a big issue.
Regarding APIs being public or not, the usual way that large open-source repos do it, is that anything that has to be used is public, but the developers will define different tiers or levels of the API. The figure here https://nightlies.apache.org/flink/flink-docs-release-2.3/docs/concepts/overview/ is a good example.
I see this being useful for ASAPPlanner but we should only do this as necessary. If its not a big concern right now, I would not do anything about it right now, since pre-mature abstraction is more likely to hurt than help.