On Wed, Jul 31, 2002 at 05:08:05PM +0300, Alexander V. Karelin wrote: > Hmm..... Hmmm........ > > Well, the problem with the first idea is that pushing stuff into stack > directly from C code is not really worthy because a simple call to "while" > may cause things to be pushed and popped around, which will break the > continium and.... Well - the parameters will be either removed or replaced > in the stack.
Wrong. If you do it properly via inline ASM, you're guaranteed that the compiler won't interfere. You have to give it the proper constraint. > As for the trick with going into stack addresses through C - it is doable, > but it is: > > 1. Illegal You mean "non portable". > 2. May cause exceptions in some cases Kindly elaborate? > 3. Is surely a bad idea for anything that has to run as a daemon or as a > server. Is a bad idea in general. But then again, so is the entire idea of calling a function in the way you want to, unless you have a very good reason. Type safety and the rest of the things the compiler checks are a heavy price to pay unless you have a very good reason. > As for the optimization suggestion of reducing the nArgs by one each time > - the suggestion is great, but my "assembler" thinking isn't good enough > yet to implement that, without doing something wrong:)))) (The new version > is on its way...) > > Anyhow, there are two additional "minor" issues: > > 1. The return code from the call (if any). Very tricky point, considering. > But it seems less complicated as it is one return value per call, and not > a variable length list of thingies. What's tricky about it? IIRC, you just put the retval in eax and be done with it. > 2. The "nArgs" has (I presume) to be replaced by an array of unsigned ints > that will specify the size of each corresponding argument. Since we may > have char and short type parameters, it is important that the code is > aware of the parameters size when pushing 'em into the stack. That's true. Of course, you could take a shortcut and always pass *ONE* argument, which is the address of a structure that has all of the other arguments, with corresponding types and sizes. Again, why do you want to do this? > Back to the safety of my c++ world that polymorphs the troubles of > assembler into an endless stream of useless instructions nobody wishes to > read...:))) Thank You. Sure, it's been an amusing excersize. -- http://vipe.technion.ac.il/~mulix/ http://syscalltrack.sf.net/
msg20807/pgp00000.pgp
Description: PGP signature
