On Tuesday 17 January 2006 16:08, Keean Schupke wrote:
John wrote:
On 2006-01-14, Keean Schupke <k.schupke@imperial.ac.uk> wrote:
Erm, has nobody replied to this yet? I want a robust interface, that uses bracket notation all the way down, so that any error is caught and resources are freed appropriately without the use of finalizers (which may not get run and lead to resource starvation - they are not reliable
To be sure, your only failure situation in this case is if you're dealing with many connections *and* creating/destroying them frequently.
Hopefully you wouldn't be.
You could be using connection pooling in the database driver or ODBC layer... Here the minimal overhead of opening/closing allows you to use a "bracket" within each connection, rather than around the whole server. Besides which the goal is not just to be safe in practice, but to be theoretically safe in all circumstances. If you allow the programmer to shoot themselves in the foot, then they often will (for example memory management and buffer overflows)... Its no good to partly remove responsibility, as that makes bugs more likely not less likely (If the programmer has to deal with an opaque system with flaws, unless the programmer is highly aware of those flaws they will take no account of them in their coding). The only way you can give the programmer a genuine black box to play with, is if it is theoretically safe, then the programmer can (ab)use it how they wish without accidentally breaking the conditions of usage.
I agree most strongly. Furthermore, as a /user/ of a library, what I want is simple semantics and guarantees are always simpler than side conditions, "take care when doing this or that", etc.. I just don't want to have to care about such stuff. Also, to make things easier one often choses bad programming style ("defensive programming") in order to avoid the complex reasoning necessary to ensure that the program "behaves well". For instance, it is tempting to encapsulate the whole program into the 'withDB' bracket, instead of only the part that actually uses the DB connection, just to be on the safe side. If the type system catches this kind of errors, one is encouraged to restrict the scope of the DB connection (or whatever), resulting in earlier freeing of resources and better modularity. Ben