Repository navigation
Continuous memory growth in jco component #350
Description
Activity
- added a commit that references this issue
on Sep 11, 2026 Hey @andreiltd was this still around? Did we need more fixes than what went into Jco? Happy to help if it's something on the Jco side that is holding this up
Hey @vados-cosmonic, I need to integrate
js-component-bindgenchanges back toComponentizeJS. See: #353.Coming from bindgen 1.x to 2.1.10 I had to come up with a workaround for "undefined task" error and created this
Embeddingadapter. I think cleaner solution would be to have something likeCallRuntimeinjs-component-bindgenthat will allow to set.call_runtime(CallRuntime::Embedding)and passFunctionBindgendirectly toabi::callskipping jco task machinery. But the adapter seems to work for now.Reacted by Victor AdossiOhhh, OK got it, thank you for the explanation! Oh yeah that would be a cleaner solution, if you want to PR it in I'm happy to take a look! I wonder how widespread those changes would be, threading the option through everywhere, probably wouldn't be too bad.
Reacted by Tomasz AndrzejakYeah, I can give it a try and see :)
Reacted by Victor AdossiReacted by Victor AdossiYeah if we can get that out fast enough we could merge & release that instead of the duplicated
Embedderlogic? I'm a bit wary/want to try the code in https://github.com/bytecodealliance/ComponentizeJS/pull/353/changes against Jco's test suite since we're overriding bindgen in a few spots.[EDIT] I can definitely get a js-component-bindgen release out today if the PR goes up!
Reacted by Tomasz Andrzejak
Repeatedly calling an export on one ComponentizeJS instance causes continuous RSS growth and shows no signs of stabilizing. Small repro can be found here https://github.com/ludfjig/jco_leak where the compoentnt exports
echo: func(value: list<u8>) -> list<u8>;and is repeatendly called.With ComponentizeJS 0.22.0:
Note that ComponentizeJS 0.19.3 does NOT repro the bug as you can see it stabilized on 188.9MiB:
I believe the bug was introduced in #319.
Note: I reported a similar bug in #279 and fixed it in #295.