On Friday, March 22, 2002, at 05:10 PM, Oleg Svirgstin wrote:
Hi,
Sorry for putting such a stupid problem to the eyes of the respected
qualified public.
Last time I wrote:
I have some problems with accessing calling "QueryInterface".
Is there an example of working with all these USB structures from C++?
"email@hidden" answered:
Are you using Apple C++ (from ProjectBuilder) or Metrowerks C++
from CodeWarrior? Is your C++ app CFM?
I am using both, MW C++ for the "Big Application" (which is CFM), and an
Apple C/C++ bundle for USB access when launched under the OSX.
First I tried wrapping USB access into MW C/C++ and after a lot of
efforts
changed the approach to separate these OSX-only calls into a mach-0
bundle.
It looks like a good solution.
Generally this approach works, I have got most of pieces of code
working for
the big app from the bundle (in plain C). Finally it is a mix of C and
C++
(with some extern "C" tricks), I tested various aspects of this mix - it
works, too.
The problem arose when I started to assemble those pieces together,
wrapping
various plain C tasks into classes/objects.
The problem (I found a working workaround, but I would love to find a
simple
correct way to do the things...) is like that (Apple C++):
// I try to get IOUSBInterface from a device reference returned from
// IOGetMatchingServices. That piece of code worked fine in plain C.
//
// Omitting details, the C++ method where I have problems looks like
this:
//
// aClass::Method( io_service_t devRef)
// {
// IOCFPlugInInterface **iodev;
// IOUSBDeviceInterface **dev;
// ...
//
// IOCreatePlugInInterfaceForService(devRef,
// CFUUIDGetUUIDBytes(...),
// &iodev, &score);
//
// err = (*iodev)->QueryInterface(iodev, CFUUIDGetUUIDBytes(...),
// (LPVOID)&dev);//here I can't compile
// ...
// }
//
The diagnostic tells "ANSI C++ forbids implicit conversions from
"void*" in
argument passing". I would be happy to know, where such conversion could
happen, since I pass arguments some of which are typed differently from
what
the plain-C interface wanted. (As far as I can see, I did not enforce
any
"strict ansi" anywhere - Apple C++ diagnostics and flags are a nice
saga of
its own.)
QueryInterface expects a void *thisPointer where I pass it a
"IOCFPlugInInterface **", but it would be rather a conversion "to void*"
than "from void*". I tried to cast iodev in various ways to "void*",
and it
drives me mad when I think there must be an easy variant I just did not
find. (Does it want a void**?)
[....]
Now the bundle starts to look like a multi-layered sandwich
(C-interface to
communicate with the big app, then C++ complex of classes, calling here
and
there another level of C routines to do some house-keeping...
[....]
Best regards
Thanks in advance for any suggestions
OIeg
I also encountered the warning about void * with IOKit and Apple (g)++.
The result I came up with is the same as yours basically - all of my
files which deal directly with IOKit types and interfaces are .c files.
If someone knows the magic compiler switch to make this stuff work
with the C++ compiler I'd also like to know. However since you're
going from PB to CodeWarrior you might be better off sticking
with C since there are potential implementation incompatibilities
between Apple C++ and MSL C++ (e.g. standard library classes)
I don't know if it's possible to avoid the layering you describe but
it may at least be possible to factor it cleanly. You may want to
gather the various IOKit types into a common data type (probably
a C struct) representing a single device or interface, then have
your module deal with that. When going to the CFM layer you can
declare an opaque (void *) type to wrap your struct so that no
one in CFM-land needs to know anything about the IOKit types.
hth,
Chris
_______________________________________________
usb mailing list | email@hidden
Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/usb
Do not post admin requests to the list. They will be ignored.
