On 11/04/2009 09:34 AM, David H. wrote:
On Nov 4, 12:00 am, Trond Norbye<[email protected]> wrote:
[...]
Other "problems" you may encounter is:
* Declarations of variables when they are used, and not in the beginning
of the scope (this includes inside for () etc)
* Some places in the code use // for comments
* uint64_t and the use of PRIu64 in printf's
* bool as a datatype
* nameless unions
* some struct initalizers where we address arrays and members directly..
Do your compiler support that?
* Declarations - not supported(!!!)
* C99 comment style - supported
* uint64_t - supported (even in C mode, I think; for sure in C++ mode,
or this would be a showstopper for me)
* bool - can be faked
* anonymous unions - supported
* offsetof(), etc. - not sure (requires more research)
Of course, the major show-stopper here are the inline declarations.
However, the more interesting question is whether we really need to
compile memcached as C code under VC9. I'm fairly sure most of these
issues will just go away if we compile as C++ code (that is, create a
VC9 project which tells Visual Studio alone to compile as C++, which
it prefers anyway). I'm sure some people will howl and holler about
exceptions, vtables, RTTI, multiple-inheritance, and a whole raft of
other completely irrelevant issues, but it seems to me that this will
make the code *just work*. Of course, I will have to play with it
quite a bit more to justify that claim. My guess is that VC9 can be
made to shut up about scary uses of struct member access and 0-length
arrays with some strategically placed warning suppression pragmas (for
instance, VC9 is quite happy to both identify 0-length array members
as "non-standard" and to allow their use in C++ mode, while it
considers them totally kosher in C mode). Is this a viable way
forward?
According to C you can assing void* to anything and from anything,
whereas C++ will generate an error if you try to do that (or can you
tweak your compiler to work around that?)...
Cheers,
Trond
