On Monday, 6 January 2014 at 09:59:55 UTC, Organic Farmer wrote:
Just found out that when I replacestruct NotNull(T) { T obj; }with (http://arsdnet.net/dcode/notnull.d)'s definition of NotNull it all makes sense.
Yes, it is very important to use the full type so you get the checks. The reason this is better than the segfault is that here, the run-time error occurs closer to the point of assignment instead of at the point of use.
my_function(enforceNotNull(obj)); // throw right here if it is null
This especially matters if the function stores the object somewhere. Having an unexpected null in the middle of a container can be a hidden bug for some time. Fixing it means finding how null got in there in the first place, and the segfault stack trace is almost no help at all. The not null things though catch it early and then the type system (almost*) ensures it stays that way.
* it is still possible to use casts and stuff to get a null in there but surely nobody would actually do that!
