Eóin: as
somebody posted at SO, it's possible to support pre-XPSP2 with VS2010, but it does require a bit mucking around... haven't looked deeply into it, but I guesstimate it's possible to have everything working, lambdas and all, without
too much trouble.
I don't agree that getting around {En,De}codePointer is a hack - and again, fscking shame on MS for not giving developers a few options here. Link.exe even refuses to set older ImageVersion fields for NO FSCKING GOOD REASON. As I don't know if there's other modifications needed to get VS2010 EXEs running on 2k, I can't comment on the hackishness of those... but I suppose it's not too bad. And hey, you get the CRT source code along with VS2010, so it should be possible to develop less-hacky solutions.
skwire: win2k supports unicode just fine, all NT builds of Windows do - and Win9x even has limited support for it via unicows.
Personally, I don't really know. For fSekrit, I develop in VS2010, but build release versions with vc2003toolkit (shame on MS for taking that offline, and double shame on them for not offering "toolkit" versions of later compilers!) so that I have win9x support... but that's the last project where I'll support win9x.
For everything else, I'm considering whether I want to go through the hoops required for win2k support. For c++ dev, I
definitely want to use the c++0x features, after having been developing in C# for a while it's just too painful not having lambdas and auto.