MonoHOM is a HigherOrderMessaging framework which doesn't use
-methodSignatureForSelector: to bounce messages, but instead the lower-level
-performv:: (specified by the ObjC runtime).
On the plus side, this means that
-reject Just Work�.
On the minus side, this means that iterating over the arguments (which you'd want to do to implement e.g.
-inject:) is a bit trickier since you're dealing with a
marg_list instead of a nice little NSInvocation instance.
http://capefish.org/monohom/trunk/ (Subversion, and browse-via-your-web-browser-if-you're-so-inclined)
Oh yeah, I'm not sure how well this will work on OS X/Intel; I don't currently have such a machine to test on, for one. For another, messages returning floating point numbers have their own
objc_msgSend variant on Intel, and this may interact with this code in "interesting" ways. Likewise I have not currently tested anything returning a
struct, for example an
NSRect, so I don't know how this will interact with the struct-returning
E-mail me (RobRix) if anything strikes you about this.
(Why "Mono" HOM? Nothing to do with the excellent mono project; it's cos my business is Monochrome Industries. I guess you could say that the framework is singular in its approach as well, but I might be pushing this a bit.)
Darnit! Why didn't I think of that! (: Seriously, mind if I rename it to that? (Depends on my copious free time whether I will or not, but still...) -- RobRix
How is it that
-reject Just Work? My understanding is that they fail because the compiler sees the BOOL return type and generates code in the caller that strips off the top three bytes. If that were the case then it wouldn't matter what the called code does. What am I misunderstanding? -- MikeAsh
I'm probably wrong, but my impression was that was not happening in compilation, but in NSInvocation/NSMethodSignature. Not using them prevents having to futz around with types, period. I'll do some more investigation, but I do trust the tests I've written, and they're (all of them) passing... -- RobRix
The compiler still has to generate code in the caller to read the type properly, so both sides have to work together. The NSInvocation approach could be worked around similarly by using an NSMethodSignature with a faked return type, but that still won't fix the code in the caller. It's possible that this problem has changed in later versions of gcc, as when I was first reading about/struggling with this problem it was way before gcc 4. I think gcc 3.1 would have been current at that time, or possibly even gcc 2.95. It wouldn't surprise me too greatly if gcc 4 no longer had this problem. -- MikeAsh
-forward:: is never called on Leopard (Mac OS X 10.5), so this HOM implementation does not work. -- AlastairHoughton