So, even if we think that “more is better”, we are faced with the question “which more”?

I recently stumbled upon your blog. I read many of the articles there, and was particularly impressed by the “Holy Trinity” article. It is beautifully crafted, as if a word-smith has gone to work.

My own background is in logic, although more recently I have been devoting increasing amount of my time to types.

Coming back to your article, I was surprised not to see complexity theory mentioned. It is our best attempt at understanding computation through the lens of cost.

I guess your reasons could have been that complexity theory itself can be interpreted using logic.

Anyways, thanks for a very inspiring post which has great clarity of thought.

-Vijay.

I agree that compositionality would be nice: But the problem is that the wildcard handlers already destroy the compositionality – or at least those aspects of it I care about. It just doesn’t seem that what’s left is important enough to focus on the secrecy.

Moreover, more often than not, at least in my code, exceptions aren’t a communication channel between two places in the code that share a binding, but rather a broadcast from one place in the code that ends up in another piece of code that was developed separately and without knowledge of the other. Providing a protocol for communication in that setting is more important than secrecy. Unfortunately, SML’s exceptions make providing a rich, shared protocol pretty difficult.

The reason for secrecy is compositionality. Say I’m a higher order function. I don’t want to accidentally intercept an exception from the function I call just because I happen to use the same name as one used by the (black box) called function. Hence the need for secrecy.

I’d characterize exceptions differently: Exceptions form a communication protocol. For this, the “shared” is much more important than the “secret”. I don’t find a compelling rationale for the “secret” either in your post or in the Book.