Rewrite more of _mega passes_ to _mini passes_

[JaroslavTulach]

This is a follow up of #10981 addressing one of the future works as provided in description of #11191- e.g. "rewrite more mini passes and do it in Java - follow SectionToBinOp example".

Goal: rewrite three+ (up to eight) mega passes into mini passes, so they can be processed by a single IR traversal.

This log indicates the current order of IR passes. The ideal ones to select are next to existing mini IR passes. For example Patterns and PrivateSymbolsAnalysis that are just next to TailCall (already rewritten by #11191). Or MethodDefinitions that is next to two already created mini IR passes.

Please rewrite the passes into Java.

Tasks

Preview Give feedback

1. Enforce no modifications to the IR elements in the preparation phase - mini passes should not attach any new metadata to the elements
2. Rename IRPass to something like IRMegaPass. Follow suggestion from Migrate some passes to Mini passes #11191 (comment)

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-06):

Progress: - #11267 PoC is finished; Interactive review and created issue #11498 so it can be scheduled

• Benchmarks in Convert Array_Like_Helpers.map to a builtin to reduce stack size #11363 are better, merging.
• Small fix for the --jvm option in --jvm tries to find Java executable system-wide. #11500
• Started to convert some more mega passes to mini passes - More IR mini passes #11501 It should be finished by 2024-11-13.

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-18):

Progress: - Finding out why there is an incosistency between ModuleContext.module.getIr and ModuleIR as passed to PassManager.

• Because of this incosistency, ImportSymbolAnalysis.createForModuleCompilation(modCtx) has modCtx.bindingsMap == null.
  But ImportSymbolAnalysis$Mini.transformModule(modIr) has modIr.passData.get(BindingsAnalysis) != null. It should be finished by 2024-11-25.

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-19):

Progress: - Fixed incosistency of IR in ModuleContext and module IR in tests. It should be finished by 2024-11-25.

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-20):

Progress: - Add tests for MiniPassTraverser - mock Expressiobn and MiniPass.

• MiniPasses can stop the traversal early.
• Migrating AmbiguousSymbolAnalysis to minipass and to Java.
• Testing mini pass chaining It should be finished by 2024-11-25.

[enso-bot]

Pavel Marek reports a new STANDUP for yesterday (2024-11-21):

Progress: - Still increasing test coverage.

• Generalize algorithm in PassManager to chain minipasses also for inline compilation. It should be finished by 2024-11-25.

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-22):

Progress: - 5 mega passes converted to mini passes.

• I will not convert any mroe passes, just focus on fixing the tests.
• Altogether, there are now 8 consecutive mini passes in globalTyping pass group.
  • This may already be visible on the benchmarks. It should be finished by 2024-11-25.

[JaroslavTulach]

there are now 8 consecutive mini passes in globalTyping pass group.

+1

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-25):

Progress: - Hunting a nasty bug in IR caches caused by this PR

• Helping Hubert with unexpected SLF4J config inside std-bits.
  • Seems like an issue with module layers?
• DevX logging rolling policy improvement request: Rolling policy of log files based on executions rather than file size #11657 It should be finished by 2024-11-25.

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-26):

Progress: - Tests are green, just need to migrate MethodDefinitions to a mini pass.

• So far, I have only converted it from Scala to Java
• For that, I need to think about how to address pass data for mini passes. It should be finished by 2024-11-28.

[JaroslavTulach]

Tests are green

Good.

address pass data for mini passes

What do you mean by that? I don't see anything special in the MethodDefinitions pass to prevent it to become mini pass.

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-27):

Progress: - Final change: Converted MethodDefinitions to mini pass.

• Tests are green, scheduling another benchmark run.
  • If no regressions are found, I will merge the PR tomorrow. It should be finished by 2024-11-28.

[enso-bot]

Pavel Marek reports a new STANDUP for today (2024-11-28):

Progress: - Final reviews integrated.

• Benchmarks are stable.
• Review of Intersection types & type checks #11600
• Starting work on Enso builtins don't expose Enso methods #11589 It should be finished by 2024-11-28.