The call method, through nogvl_context_call, does not seem to enforce the max_memory limit

[masongup-mdsol]

I am still investigating what's happening here, but it appears that the max_memory limit is only enforced when running Javascript code through the exec method - see the code setting the data point on the isolate and the AddGCEpilogueCallback here, in the nogvl_context_eval method. No comparable code seems to be present in the nogvl_context_call method.

The code that I am currently working with loads a few hundred lines of JS via the eval method which defines various Javascript functions, and then calls the defined methods with Ruby objects using the call method. I was hoping to use the max_memory limit in this code to prevent excessive memory consumption, but it does not seem to be effective. I added a log line to the C callback function gc_callback, and the memory limits I set appear to get stomped by the time I run any of the defined functions - when it runs, it thinks that the softlimit is some random huge value.

At this stage, what I am asking is whether there is an intentional reason for this. If this is not an intended behavior, I am open to creating a test exercising this, and writing a fix.

[SamSaffron]

I don't think this is intentional, would be more than happy to see a fix.

[masongup-mdsol]

I've been looking into this some more. Based on what I've seen, it seems that something is overwriting the value for the MEM_SOFTLIMIT_VALUE in the isolate under some conditions. When I tried to write a Ruby test for it in mini_racer, it was overwritten to 0, and would exit on the first GC run. When I was running the original code in the application I'm working, it would get overwritten to a very high value, and would never trigger. I see 2 possible ways to address this issue:

1. Figure out and fix whatever is overwriting the MEM_SOFTLIMIT_VALUE. I am not an expert in C++, and probably don't have bandwidth to figure that out.

2. Add a bit of code in nogvl_context_call to set the memory limit if present, just like is already present in nogvl_context_eval. I have implemented this locally and tested it, and it seems to work as expected, both in the test I wrote on mini_racer, and in our application. I'll open a PR with this change.

One other note - it doesn't seem to be possible to write a Ruby test exercising this issue. The memory limiter still runs with or without the fix, it just runs with an incorrect value for MEM_SOFTLIMIT_VALUE, and thus terminates at the wrong time.

[masongup-mdsol]

This has been fixed with the linked PR #161, and released in version 0.2.10. Thanks for the support!