[meeting] [joint] 28.06.2024

[deemp]

Nickolay's status

Plans

1. Normalization with delayed substitutions (Experiment with closure of objects #421)

   • This approach should significantly improve the normalizer speed on big examples.
   • Can't completely debug locally yet due to problems with ρ.
   • The work will be continued.

2. Custom rules (Custom rules. Iteration 2 #423)

   • @fizruk will hopefully implement the rules by the meeting next week

Questions

Optimizing streams (streamliner) - open questions for @yegor256 @maxonfjvipon:

1. Did you try to run the prototype (streamliner)? What were the results?
2. What are the downsides of streamliner?
3. The streamliner doesn't support parallel stream pipelines (stated in the article). How important is this issue in practice?
4. Which Java versions should we support?
5. According to the article, "The key to be able to inline the relevant calls ahead-of-time is precise type information to enable virtual call resolution." Hence, the usage of EO and PHI may be non-optimal or impossible because the information about types is lost.
   What are strong reasons to use EO and PHI for such optimizations?

[deemp]

@yegor256, @maxonfjvipon, please answer the questions

[maxonfjvipon]

@deemp
1-2. we didn't try streamliner for now, maybe later, but we definitely won't use it.
3. for now it's not really important, I believe
4. I think it would be good to support all LTS versions: 8, 11, 17
5. After translation to PHI we will have the info about types, you may check how it would look like here so it won't be lost. We want to use EO and PHI because we just want to use it) Optimizing Streams is just a first step, first optimization we want to introduce. In the future there will be more.