Avoid AliasAnalysis for Hello_World.enso

[JaroslavTulach]

Pull Request Description

Fixes #10843 in an even better way then #10906. At the end of work on #10906 it has been found that AliasAnalysis is still used for Hello_World.enso. Mostly because when loading its .ir cache, we cannot resolve its Standard.Base.IO import as nobody loaded the Standard.Base package into the system yet. This PR fixes the problem by delaying resolution of BindingsMap internals until one of its getters is called.

At the end Hello_World loads its .ir caches properly without invoking any AliasAnalysis functionality as verified by checking the logs.

Checklist

Please ensure that the following checklist has been satisfied before submitting the PR:

• All code follows the
  Scala,
  Java,
• Unit tests have been written where possible.
• Benchmarks are improved - 1st run - results are good

Steps to Reproduce

The ultimate way to verify the behavior is in debugger. Place a breakpoint to Graph.scala:12 - it should get trigger only once when LocalScope.empty is being created. Possibly also for the second time when execution is over and #10842 kicks in (this second case can be disabled by using --repl for example).

To reproduce from CLI run twice (first run generates the .ir cache) and then checks the logs:

enso$ ./built-distribution/enso-engine-*/enso-*/bin/enso --run test/Benchmarks/src/Startup/Hello_World.enso
enso$ ./built-distribution/enso-engine-*/enso-*/bin/enso --run test/Benchmarks/src/Startup/Hello_World.enso --log-level debug | grep Hello

prior to this PR the log contains:

[enso.org.enso.interpreter.runtime.SerializationPool] Unable to load a cache for module [Hello_World].
[enso.org.enso.interpreter.runtime.SerializationPool] Deserializing module Hello_World from IR file: false
[enso.org.enso.compiler.Compiler] Loading module [Hello_World] from source.
[enso.org.enso.compiler.Compiler] Generating code for module [Hello_World].

with this PR the log contains:

[enso.org.enso.compiler.Compiler] Parsing module [Hello_World].
[enso.org.enso.interpreter.runtime.SerializationPool] Restored IR from cache for module [Hello_World] at stage [AFTER_STATIC_PASSES].
[enso.org.enso.interpreter.runtime.SerializationPool] Deserializing module Hello_World from IR file: true
[enso.org.enso.compiler.Compiler] Generating code for module [Hello_World].

[JaroslavTulach]

This is the core of the change. Delaying toConcrete until all modules are registered to package repository.

It is also potentially dangerous change. Previously, when the restoreFromSerialization failed, it would return None and that would trigger parsing. Right now we optimistically claim everything is all right and then we may get failures later when someone calls for example ResolvedModule.unsafeModule.

It is probably OK (everything is green; but we aren't testing broken scenarios much), but we should get ready for such reports in the future and fix them when they appear.

[JaroslavTulach]

@hubertp, when this exception is logged, the stack trace isn't printed on the console. What's the recommended way of logging an exception with stacktrace via context.logSerializationManager?

[hubertp]

I don't think I'm surprised anymore. It looks like the included application.conf does not redefine the layout of logging, meaning it does not print exceptions.

[hubertp]

I will file a separate ticket to enable those. Right now the default and only appender is console. We could easily log those to the file but I'm still against printing those in the console.

[JaroslavTulach]

There is 5% speedup on Import_World startup benchmark:

obviously Hello_World startup shows lower improvements (around 100ms) and empty_startup shows none (as there is no import). Trying yet another run and the results are also good.

[JaroslavTulach]

I don't trust local benchmarking, but per @Akirathan request I did one:

Old Run

Benchmark                     Mode  Cnt     Score      Error  
Startup.empty_startup         avgt    3  2438.773 ± 2213.236  
Startup.hello_world_startup   avgt    3  3550.110 ±   65.246  
Startup.import_world_startup  avgt    3  4118.481 ±  938.980 

New run

Benchmark                     Mode  Cnt     Score      Error 
Startup.empty_startup         avgt    3  2468.019 ± 1609.600
Startup.hello_world_startup   avgt    3  3323.337 ±  811.669
Startup.import_world_startup  avgt    3  4097.680 ±  569.984

The benchmark fluctuates a lot, but there seems to be no slowdown and possibly there is some speedup.

[hubertp]

Minor Scala style issues

[hubertp]

I will file a separate ticket to enable those. Right now the default and only appender is console. We could easily log those to the file but I'm still against printing those in the console.

[enso-bot]

Jaroslav Tulach reports a new STANDUP for yesterday (2024-09-06):

Progress: .

• Avoid AliasAnalysis ready for review: Avoid AliasAnalysis for Hello_World.enso #10996
• approved Oracle GraalVM fix: Remove module that was causing failures for enterprise edition of GraalVM #10991 (review)
• folding benchmarking output ready for review: Fold output of each benchmark #10914
• equality of MetadataStorage: Two IR objects aren't the same after de-serialization #10985 (comment)
• preparing for meetup: Insight script demo
• libs standup It should be finished by 2024-09-10.

[Akirathan]

LGTM. Minor nitpicks.

[enso-bot]

Jaroslav Tulach reports a new STANDUP for yesterday (2024-09-09):

Progress: .

• AliasAnalysis merged: Avoid AliasAnalysis for Hello_World.enso #10996
• closed Remove AliasAnalysis metadata from LocalScope #10833
• Fix Comparator warnings should propagate to output of == #10679 by Propagate comparator warnings via Any.== #11009
• picking up work on Two IR objects aren't the same after de-serialization #10985
• meeting Dmitry (IR traversal - Avoid IR traversal by converting (some of) our Passes to mini passes #10981)
• folding merged: Fold output of each benchmark #10914
• PR for equality: Enabling structural equality for MetadataStorage again #11012
• enabling structural equality: Enabling structural equality for MetadataStorage again #11012
• solving StackOverflowError: Enabling structural equality for MetadataStorage again #11012 (comment)
• remove are they same check on release branch: Disable check for equals when two objects collide #11015
• debugging annotation processor It should be finished by 2024-09-10.