February 2011
« Jan   Mar »





One of my fellow lab-mates has already posted about const-correctness, because of some similar issues. My work today, takes it a tiny bit further into sheer insanity.

If you don’t start your program off using a const-correctness discipline, it’s very likely that you will find yourself in a world of hurt when you try to add it in later. I hit something like this today. In our system we have these labels, which are managed by a Factory and which are supposed to be immutable. Because all the labels are managed by a factory, we have a whole bunch of pointers:

JSFlowLabel *lab = ...

Although no bugs have come up as a result, we should have decided a long time ago, to enforce the immutable semantics with the compiler. So, after seeing a few suspicious lines today, (which turned out to be fine, but which raised my paranoia alarms) I decided that we should add const. Beginning with the Factory that produces these:

JSFlowLabel const* JSFlowLabel::create(...) {...}

Followed by a hundred changes to code like the above, and a few method call modifications.

JSFlowLabel const *lab = ...
JSFlowLabel::join(JSFlowLabel const *lab1, JSFlowLabel const *lab2) {...}

But, of course, just when you think you are nearing completion, there’s some line lurking somewhere to bite you.

js::Value val = ...
JSFlowLabel const *lab = ...

And what, pray tell, is the signature of setPrivate(...) ?

void setPrivate(void *data);

Oh, well that’s easy to fix, I think to myself. I’ll just overload a const version and dispatch.

void setPrivate(void const *data) {
    setPrivate(const_cast<void *>(data));

Having got this far, I still have some hope things will work out. I start the compiler to test my changes, and wonder a bit about what deep ruminations about data have led the compiler to draw such a sharp philosophical distinction between void and const void. And despite these wonderful distinctions, I’m still never allowed to make data of type void, only functions that can return a void whatever that means, … wait, can I have a function returning a const void? If I continue this line of thought, will I become as useless as a philosopher? Does enough rumination about void type your thoughts as void? Is the brain-rot setting in? Will I ever be able to recover the damage C++ has caused me?. Oh, yeah, back to my story…

So, I was thinking things might work themselves out, and (at worst) I’d have to make a few more wrapper overloads to some system-level functions. But NO! my very sanity was soon challenged by the compiler:

../jsfun.cpp:1158: error: call of overloaded ‘setPrivate(NULL)’ is ambiguous
../jsobj.h:742: note: candidates are: void JSObject::setPrivate(void*)
../jsobj.h:747: note:                 void JSObject::setPrivate(const void*)

At this point my all my thoughts were dumped to /dev/null. The C++ typing system is clearly more depraved than I am, so it wins this round!

Leave a Reply