test: increase change_in_bytes for napi_adjust_external_memory test #57351
Conversation
JavaScriptCore ignores small deltas in its interface for reporting external memory: https://github.com/WebKit/WebKit/blob/8bbba2aa777cf326e24a4e5c63b1f51bd52315ff/Source/JavaScriptCore/heap/HeapInlines.h#L213-L229 More broadly, it seem reasonable for any other JS engine to do the same. Passing a large change ensures that the change is actually applied and the new adjusted value will be greater than zero (otherwise, the test relies on some external memory previously being reported as allocated). In turn, this makes the test more applicable to non-V8 implementations of Node-API.
nodejs-github-bot
added
needs-ci
| @@ -215,7 +215,8 @@ static napi_value testAdjustExternalMemory(napi_env env, napi_callback_info info | |||
| napi_value result; | |||
| int64_t adjustedValue; | |||
|
|
|||
| NODE_API_CALL(env, napi_adjust_external_memory(env, 1, &adjustedValue)); | |||
| // Some JavaScript engines may ignore small increments passed to napi_adjust_external_memory | |||
There was a problem hiding this comment.
Can you be more specific? What is the constraint you're seeing?
There was a problem hiding this comment.
For JSC, right now the cutoff is set to 256 bytes: https://github.com/WebKit/WebKit/blob/8bbba2aa777cf326e24a4e5c63b1f51bd52315ff/Source/JavaScriptCore/heap/Heap.h#L645
I realize another possible solution is that implementations with imprecise external memory tracking could treat change_in_bytes as a floor on the return value -- basically, pretending that at least the currently requested change has had an effect. On Node's side this would basically be a docs change instead of a test change. At a glance from the PR #14310 where this was originally added, it seems like the question of what to do with an engine that cannot track external memory at all was brought up but not resolved, and it doesn't look like the possibility of memory tracking existing but with a minimum was brought up. The documentation only really describes this function's effect in V8.
I'd be happy with either approach.
There was a problem hiding this comment.
IMO, the node-api adapter for WebKit should normalize small values if this is a WebKit constraint.
JavaScriptCore ignores small deltas in its interface for reporting external memory:
https://github.com/WebKit/WebKit/blob/8bbba2aa777cf326e24a4e5c63b1f51bd52315ff/Source/JavaScriptCore/heap/HeapInlines.h#L213-L229
More broadly, it seem reasonable for any other JS engine to do the same. Passing a large change ensures that the change is actually applied and the new adjusted value will be greater than zero (otherwise, the test relies on some external memory previously being reported as allocated). In turn, this makes the test more applicable to non-V8 implementations of Node-API.