Based on a recent discussion that was left out of scope of #128, there was an idea to make ->parent() return a Node rather than require callers to use shared_from_this (or intrusive_ptr_from_this now).
This PR makes that change. It:
changes parent() to return Node
adds parent_unsafe() that keeps old behavior if you don't want refcounting overheads and you know your borrow is safe
Point of discussion: what do we do with original calls to parent? Do all those calls in YAML change type, or do we make them parent_unsafe? I'm told it might have performance implications. So far, I only made ->parent()->intrusive_ptr_from_this() into ->parent() and otherwise went the unsafe route, but there's plenty of room for "don't care keep it simple" in other cases.
There's also a quick fix here for the test cases breaking in dependent projects, since this whole conversation is part of updating rego-cpp's dependencies. If this takes too long I can split that out. Moved to another PR
Based on a recent discussion that was left out of scope of #128, there was an idea to make
->parent()return aNoderather than require callers to useshared_from_this(orintrusive_ptr_from_thisnow).This PR makes that change. It:
parent()to returnNodeparent_unsafe()that keeps old behavior if you don't want refcounting overheads and you know your borrow is safePoint of discussion: what do we do with original calls to
parent? Do all those calls in YAML change type, or do we make themparent_unsafe? I'm told it might have performance implications. So far, I only made->parent()->intrusive_ptr_from_this()into->parent()and otherwise went the unsafe route, but there's plenty of room for "don't care keep it simple" in other cases.There's also a quick fix here for the test cases breaking in dependent projects, since this whole conversation is part of updating rego-cpp's dependencies. If this takes too long I can split that out.Moved to another PR