Where should the type information be: in tags and descriptors 428

Usually there will be an operator that will let you get at the bits of any type to inspect them. When generating numbers, usually one is allowed to put numbers together as one wishes.

This is a problem. Perhaps solvable via extensible systems such as in XML. If Iliffe's rules are adopted for safety, you cannot perform arithmetic operations on addresses. A special operation that selects a field from a region or transforms the region to one of the same size or smaller can be used. See the basic language machine book previously referenced for details. Note that the OS would have a master descriptor that permits description of all storage. Similarily, jump addresses or control objects cannot be changed. If the descriptor that the processor uses to describe the page table is "number" then there is no problem. This would be a no-no as described above. This can be a serious problem. Most tagged architectures write the tags and descriptors to disk (one way or another). It used to be that the extra storage required was simply prohibative. I don't believe that this would be a particular issue now, especially as more and more serial I-O architectures are used. Note that more and more data is now transferred across the network in self-identifying formats: XML, CORBA, and Sun's XDR come to mind. The ability to use instructions for generating code can be limited as was done on the Burroughs. Of course this can be gotten around by persons who have access to raw hardware. Part of the advantage of Tagged Architectures is that some level of debugging is always present. An exception error will be raised on many of the more standard problems: trying to access outside the buttigned stack area or outside a buffer. Trying to jump into data or into a procedure at other than an entry point. Trying to illegally construct an address, ... This catches many errors which are caused when someone enters data that was not planned for.

Good question. I hope that someday, someone will take it on. I suspect it depends very much on what tags are implemented and how as well as the operations that are implemented and the degree of parallelism that can be achieved against a problem. See for example S.S. Reddi's and my paper on restructurable computer architecture. Ed

----== Posted via Newsfeeds.Com - Unlimited-Uncensored-Secure Usenet News==---- ----= East and West-Coast Server Farms - Total Privacy via Encryption =----