Path: utzoo!attcan!uunet!husc6!uwvax!tank!uxc!uxc.cso.uiuc.edu!m.cs.uiuc.edu!grunwald From: grunwald@m.cs.uiuc.edu Newsgroups: comp.lang.c++ Subject: explicit member initialization Message-ID: <4800043@m.cs.uiuc.edu> Date: 3 Nov 88 17:36:00 GMT Lines: 38 Nf-ID: #N:m.cs.uiuc.edu:4800043:000:1815 Nf-From: m.cs.uiuc.edu!grunwald Nov 3 11:36:00 1988 One of the disappointing complexities of C++ is that you don't get very much control when initializing members that are class instances, as well as the previously mentioned problem of specifying the ordering of global member initializations. Certainly, you can say: Foo::Foo(int x, char y) : instance(x,y), another(x*y) and the like, but realistically, there's no reason you couldn't delay the creation of an instance and the call to the constructor. In certain cases, this can be invaluable, e.g., you need to compute something before initializing a member. As it is, you have two choices: create it using ``new'' or have some way to ``reset'' a class. E.g., I have a class MLCG that implements a random number generator, and I have a subclass of Thread allocates a random number generator based on the ID number of the thread. The thread needs to reserve a global lock, grab a counter, release the lock and then initialize the MLCG instance. Right now, I have two options: (1) have the parent of a thread compute the information (2) ``initialize'' the MLCG structure twice, once using a default and once using the values I want. Option (1) is fine, except that you wind up passing a lot of information to threads that maybe you don't care about. Also, you need to pass it from the point you say ``new Thread'', which is the wrong place to compute such Thread-specific information. Also, it's faster to have 10 CPUs executing threads that are initializing themselves than to have 1 CPU executing a thread that initializes everyone else. With data-flow information from the compiler, you can certainly issue warnings like ``use of XYZ before constructor called'', or better yet, just force a call to the constructor at that point. The only problem is that there isn't a syntactic mechanism for this.