GROOVY-12138: emit invokedynamic call sites for property writes (spike)#2674
Open
paulk-asert wants to merge 2 commits into
Open
GROOVY-12138: emit invokedynamic call sites for property writes (spike)#2674paulk-asert wants to merge 2 commits into
paulk-asert wants to merge 2 commits into
Conversation
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## master #2674 +/- ##
==================================================
- Coverage 68.7019% 68.6439% -0.0580%
- Complexity 34123 34126 +3
==================================================
Files 1536 1536
Lines 128912 129018 +106
Branches 23369 23400 +31
==================================================
- Hits 88565 88563 -2
- Misses 32517 32619 +102
- Partials 7830 7836 +6
🚀 New features to boost your workflow:
|
Property writes were the last un-cached dispatch surface: even in indy mode they compiled to static ScriptBytecodeAdapter.setProperty calls, paying getMetaClass() plus a full MetaClassImpl.setProperty resolution walk on every write, in both indy and classic modes. Behind the compile-time flag groovy.indy.setproperty (default: off), AsmClassGenerator now diverts plain (non-safe, non-spread) property writes to a new CallSiteWriter.makeSetPropertySite hook. The default implementation is the classic adapter call, so classic mode and flag-off builds are unchanged; IndyCallSiteWriter overrides it to emit invokedynamic setProperty:(ReceiverType, ValueType)V with the receiver first, preserving the runtime's arguments[0]-is-receiver invariant. The new SetPropertySelector (replacing the 'not supported' GroovyBugError for CallType.SET) fast-paths only shapes whose resolution is provably sender- and metaclass-independent: a public setter whose backing field is private or absent (GROOVY-8283 field-precedence is sender-dependent), or a public non-final field, each accepting the runtime value type directly. Everything else — Maps (GROOVY-8065/11367), Class receivers, categories, overridden/mixed-in/expando setProperty (detected by both a reflection probe and a metaclass pickMethod probe; ExpandoMetaClass is excluded wholesale, mirroring the GET selector), value coercion, listeners, propertyMissing — binds the sender-aware ScriptBytecodeAdapter.setProperty as the call-site handle, making non-fast-path behavior identical to the classic path by construction. Guards, PIC, and hit-count promotion come free from the existing call-site machinery; the value participates in the argument-class guards so type changes at a site re-select correctly. Measured: ~42x on a hot property-write loop (6M mixed setter/field writes: ~435ms -> ~10ms, written state consumed and verified). Full root test suite (15678 tests) passes with the flag enabled for all test-code compilation; a 12-scenario semantic stress produces identical output compiled with and without the flag.
blackdrag
reviewed
Jul 10, 2026
| // otherwise (no meta property, listeners, expando, propertyMissing, ...): adapter path | ||
| } | ||
|
|
||
| private static final Class<?>[] SET_PROPERTY_PARAMS = {String.class, Object.class}; |
Contributor
There was a problem hiding this comment.
just for style, this should be before the methods. Nothing blocking in my eyes
| MethodHandle sba = MethodHandles.insertArguments(SBA_SET_PROPERTY, 1, sender); | ||
| sba = MethodHandles.insertArguments(sba, 2, name); // -> (Object value, Object receiver)void | ||
| handle = MethodHandles.permuteArguments(sba, | ||
| MethodType.methodType(void.class, Object.class, Object.class), 1, 0); // -> (receiver, value)void |
Contributor
There was a problem hiding this comment.
I would consider not using SBA directly here. Instead I would make a method maybe even in the the selector, that has the arguments already in the right order and allows you to use a single insertArguments to inject sender and name and has the right remaining signature for the invocation. For example (name, receiverClass, receiver, value). Then one insertArguments call would be enough. You can still let that one call SBA... though I imagine we want to inline the remaining logic there in the long run.
Contributor
Author
There was a problem hiding this comment.
changed if I understood you correctly
✅ All tests passed ✅🏷️ Commit: 9ce62e0 Learn more about TestLens at testlens.app. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Property writes were the last un-cached dispatch surface: even in indy mode they compiled to static ScriptBytecodeAdapter.setProperty calls, paying getMetaClass() plus a full MetaClassImpl.setProperty resolution walk on every write, in both indy and classic modes.
Behind the compile-time flag groovy.indy.setproperty (default: off), AsmClassGenerator now diverts plain (non-safe, non-spread) property writes to a new CallSiteWriter.makeSetPropertySite hook. The default implementation is the classic adapter call, so classic mode and flag-off builds are unchanged; IndyCallSiteWriter overrides it to emit invokedynamic setProperty:(ReceiverType, ValueType)V with the receiver first, preserving the runtime's arguments[0]-is-receiver invariant.
The new SetPropertySelector (replacing the 'not supported' GroovyBugError for CallType.SET) fast-paths only shapes whose resolution is provably sender- and metaclass-independent: a public setter whose backing field is private or absent (GROOVY-8283 field-precedence is sender-dependent), or a public non-final field, each accepting the runtime value type directly. Everything else — Maps (GROOVY-8065/11367), Class receivers, categories, overridden/mixed-in/expando setProperty (detected by both a reflection probe and a metaclass pickMethod probe; ExpandoMetaClass is excluded wholesale, mirroring the GET selector), value coercion, listeners, propertyMissing — binds the sender-aware ScriptBytecodeAdapter.setProperty as the call-site handle, making non-fast-path behavior identical to the classic path by construction. Guards, PIC, and hit-count promotion come free from the existing call-site machinery; the value participates in the argument-class guards so type changes at a site re-select correctly.
Measured: ~42x on a hot property-write loop (6M mixed setter/field writes: ~435ms -> ~10ms, written state consumed and verified). Full root test suite (15678 tests) passes with the flag enabled for all test-code compilation; a 12-scenario semantic stress produces identical output compiled with and without the flag.