On Mon, Feb 4, 2008 at 6:06 AM, Yitzchak Gale
Simon Marlow wrote:
However, I do agree that exceptions should generally be used for exceptional conditions, rather than for general control-flow. This is an example of a *good* reason to avoid an exception: because to use an exception for a non-exceptional condition is poor style.
That is also my opinion. Some people have disagreed. So this should be the focus of the decision then.
I also agree that exceptions should only be used for exceptional conditions; but I do think that this is one of those situations. I'll try to explain why below. (Thanks, Yitzchak, for forking off the thread about fixing limitations of MonadIO.) I assume that the following conditions are already considered exceptional (I haven't ever heard anyone complain that they throw an exception): - readFile : read a nonexistent file - writeFile : write to a file in a nonexistent directory - getEnv: retrieve an unset environmental variable - System.Console.Readline.readInitFile: read a nonexistent .inputrc file I think of readHistory and writeHistory as analogues to the above functions. So calling writeHistory when the directory doesn't exist should produce an exception. And if you call readHistory, either check first that the file exists, or else expect that an exception will sometimes be thrown (just as is already the case for readFile). Packages like shellac-readline can wrap the two steps of "only readHistory if file exists" in a higher-level abstraction; but System.Console.Readline itself should just be a thin binding to the readline APIs. All that being said, this is a relatively minor issue, so if many people are strongly opposed to throwing an exception even after what I've said above, I'll accept the consensus decision. Barring that, however, I'm in favor of the interface already used by other, similar functions in the libraries. Thanks, -Judah