Peter Dimov wrote:
Johan Råde wrote:
John Maddock wrote:
That one is problematic as it really would make the array a non-POD so initialisation would occur during the dynamic phase rather than the static phase.
The compiler is allowed to statically initialize anything it can, and in this case VC 7.1 does.
True. Although there are other benefits that POD'ness brings: union support, pass through elipsis etc. However, since those are less useful, I might be persuaded that the fix is OK after all if VC7.1 really does the right thing by static initialisation.
IMO that would be a deal braker :-(
We aren't actually testing static array initialization, are we? How do we know that it works at all? :-)
Um, good question. It is very important though.
So an array class with a non-template assignment op is an aggregate but is not a POD. This language sometimes drives me nuts ;-)
Non-POD aggregates are common, consider
struct X { std::string a, b; };
X x = { "a", "b" };
Absolutely, and I've used boost::array with non-POD types as well as POD's. However, the ability to have POD's statically initialised "just like a C array" is a very useful one. John.