[ammarahm-ed]

Every Java object returned to JS stayed alive for the life of the runtime. Its wrapper was held by a strong reference in m_idToObject, and a host-object proxy created for the wrapper pinned it a second time, so returning a lot of objects grew the process until Android killed it. This PR makes the wrapper the only thing that keeps a returned object alive, and ties the Java side to the wrapper's lifetime. It changes the Node-API and the direct runtime the same way.

• Weak wrappers. A wrapper for a Java-returned object is held by a weak reference. Instances constructed from JS (extend, interface implementations) stay strong, because Java may call back into them at any time.
• No proxy for plain instances. Only Java arrays keep a host-object proxy, for indexed access. Any other returned object is the wrapper itself, and an object that already has a proxy still returns that proxy so identity is preserved.
• Native pressure. Each weak wrapper reports about 1 KB to the engine (js_adjust_external_memory / EngineHost::AdjustExternalMemory), so the engine schedules a GC under native pressure instead of only when its own heap grows.
• Release on collection. When the engine collects a wrapper, its finalizer queues the release to the post-GC drain, which marks the Java instance weak, drops the id entry and returns the memory credit. In the direct runtime the wrapper's native state destructor does the queueing, through an OwnerToken that OnDisposeRuntime clears, because the engine can destroy native state after the manager is gone.
• Recreated wrappers. Java reuses the id of an instance it has made weak, so EnsureInstanceStrong moves it back to the strong table whenever a wrapper is (re)created. If a new wrapper was linked under an id between the collection and the drain, the drain leaves that id alone.

Results

V8 release build on a Sony SO-01M (Android 11, arm64), AOT-compiled with cmd package compile -m speed. One warm-up launch is discarded, then 5 launches; values are the median ns per call.

Case	Node-API before	after	Δ	Direct before	after	Δ
Return a new object	28,339	12,744	-55%	19,349	12,359	-36%
Call a method on a new object	35,895	19,368	-46%	28,239	19,891	-30%
Return a Date Object	28,710	13,411	-53%	20,413	13,814	-32%
Return this	1,338	1,261	-6%	1,312	1,354	+3%
Return an Int Array	27,547	27,918	+1%	18,232	20,068	+10%
Native heap after run (MB)	268 MB	106 MB	-60%	173 MB	97 MB	-44%

The heap row is the native heap after one pass of the suite, which returns about 200k objects. Rows this change does not touch moved by up to ±17%. Invoke string callback on the direct runtime is bimodal in this build (7.9k to 11.8k ns across the five runs, against 7.8k to 8.4k before); it does not go through this code for its arguments, and the Node-API runtime is flat on it, so I am treating it as noise, but I have not isolated it.

All cases

Case	Node-API before	after	Δ	Direct before	after	Δ
Void Method on instance	631	662	+5%	584	596	+2%
Field on instance	807	804	-0%	894	835	-7%
Int Field on instance	365	377	+3%	395	386	-2%
Get Static Field	836	813	-3%	856	850	-1%
Void Static Method	309	292	-6%	299	313	+5%
multiply(a, b)	478	486	+2%	514	520	+1%
Pass and return string	1,154	1,218	+6%	1,164	1,202	+3%
Return a String	821	836	+2%	746	760	+2%
Return an Int	330	315	-5%	306	315	+3%
Pass a String	660	700	+6%	654	687	+5%
Return a Boolean	316	313	-1%	296	313	+6%
Return a Double	392	399	+2%	388	395	+2%
Pass an Int	393	415	+6%	407	420	+3%
Pass a Double	389	397	+2%	412	425	+3%
Pass a Boolean	476	480	+1%	432	424	-2%
Return an Int Array	27,547	27,918	+1%	18,232	20,068	+10%
Pass an Int Array	3,051	3,264	+7%	3,983	4,135	+4%
Return a String Array	27,902	29,130	+4%	20,155	21,210	+5%
Pass a String Array	5,354	5,661	+6%	6,641	6,846	+3%
Return a Double Array	26,756	27,120	+1%	19,726	20,490	+4%
Pass a Double Array	3,049	3,134	+3%	3,901	3,970	+2%
Return a Boolean Array	25,451	26,139	+3%	18,996	19,611	+3%
Pass a Boolean Array	2,870	2,980	+4%	3,660	3,812	+4%
Int Array[0] read	348	361	+4%	245	240	-2%
String Array[0] read	777	784	+1%	671	666	-1%
Double Array[0] read	465	426	-8%	336	342	+2%
Boolean Array[0] read	339	350	+3%	234	234	+0%
Int Array[0] write	476	511	+7%	360	364	+1%
String Array[0] write	1,163	1,220	+5%	1,035	1,072	+4%
Double Array[0] write	468	546	+17%	374	373	-0%
Boolean Array[0] write	464	513	+10%	369	367	-1%
Return a Date Object	28,710	13,411	-53%	20,413	13,814	-32%
Pass a Date Object	3,853	3,688	-4%	3,656	3,749	+3%
Invoke void callback	7,647	7,358	-4%	7,214	6,899	-4%
Instance add(a, b)	840	832	-1%	806	796	-1%
Pass and return non-ASCII string	1,161	1,252	+8%	1,189	1,190	+0%
Pass and return 1KB string	5,166	5,040	-2%	5,064	5,114	+1%
Return a Char	351	343	-2%	341	350	+3%
Return this	1,338	1,261	-6%	1,312	1,354	+3%
Return a new object	28,339	12,744	-55%	19,349	12,359	-36%
Call a method on a new object	35,895	19,368	-46%	28,239	19,891	-30%
Pass a 16 Int Array	4,138	4,668	+13%	4,935	5,262	+7%
Pass a 16 Double Array	4,803	5,012	+4%	5,761	5,931	+3%
Invoke int callback	10,298	10,765	+5%	12,617	10,634	-16%
Invoke string callback	8,837	8,908	+1%	8,125	11,360	+40%
Native heap after run (MB)	268 MB	106 MB	-60%	173 MB	97 MB	-44%

Only V8 was built and run.

Jasmine suite

Run once on a debug build of the top of this stack (this PR, the bulk array reads in the PR above, and the crash-handler PR), on the same device:

	main	stack top
Node-API	523 specs, 0 failures	522 specs, 0 failures
Direct	451 specs, 1 failure (triggers interval, a timing spec)	450 specs, 0 failures

The spec count is one lower on the stack top because the crash-handler PR removes a disabled spec. This PR was not run on its own.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: e52267ab-e946-4c23-8459-6a7d82c111fc

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

• X
• Mastodon
• Reddit
• LinkedIn