Feature #13133: TracePoint: Add event type for constant access

• Status: Rejected
• Assignee: ko1 (Koichi Sasada)
• Target version: -

[burke (Burke Libbey)]

Hi there,

I've attached a patch to add a new :constant_access TracePoint event.

One feature and/or bug, depending on your perspective: since this event fires from setinlinecache, the event is only triggered:

• after nested constant resolution: for A::B, only one event is triggered (because getinlinecache/getconstant/getconstant/setinlinecache);
• on cache miss: it cannot be used for things like counting iterations, which limits possible uses, but reduces overhead (because getinlinecache jumps over setinlinecache on a hit).

The use-case I have in mind here is to enable building runtime package/component boundary enforcement systems. For this case, both of the above properties are ideal.

Enforcing package boundaries is a problem we've been discussing a lot lately at Shopify.

If you want a more concrete usage example for this patch, I have some proof-of-concept code posted that makes use of it to construct a fairly minimal package system. Packages::CONSTANT_ACCESS_TRACEPOINT is the particularly relevant bit.

[shyouhei (Shyouhei Urabe)]

• Status changed from Open to Assigned
• Assignee set to ko1 (Koichi Sasada)

[ko1 (Koichi Sasada)]

Enforcing package boundaries is a problem we've been discussing a lot lately at Shopify.

I understand the issue. But your patch and your design needs more consideration.

• Your patch provide a hook only on const cache miss, but the name constant_access I assume it hooks all of constant accesses.
• Your patch inserts the hook only in insns.def, but C level constant access is available.
• Not only constant access, constant setting or reopen class/modules are also needed to analyze (and introduce restriction if needed into ) program behavior.

[burke (Burke Libbey)]

All good points, thank you. I definitely agree that the implementation is not good. Feel free to close this. I'll see if I can find a better way to accomplish my goals here some other time.

[ko1 (Koichi Sasada)]

• Status changed from Assigned to Rejected