wagner (02/01/83)
(this appeared originally in net.micro.68K, but is copied here since it has more general applicability. The two versions are almost identical, except for this note; it isnt worth reading both. Sorry) I was looking at small C v2 to solve some problems for me as a standard amongst my various machines and amongst my various friends machines in the neighbourhood. I am trying now to get it up on UNIXtm as a first step to porting it to my 6809, and trying is the right word. The language it is written in is close enough to C to fool the unaided eye, but it sure dont fool the compilers. I dont mind a subset C compiler, but I mind a compiler that accepts syntax outside the language (not extentions, just bugs). The compiler itself misuses stdio so badly that it wont compile with UNIX stdio. Some examples: 1) the files are all declared int instead of FILE *. The solution appears to be to change them all to FILE *, and then, in the header files, #ifndef UNIX #define FILE int #endif and then change his support routines to expect *int instead of int. Eventually, of course, he should be passing more info than just an int around if he wants to do things right, but at least this gives him the option of leaving the shape of a FILE "variable" to the stdio.h file. 2) He tries in several places to read from stderr. Needless to say, UNIXtm isnt amused. 3) He is forever assigning pointers to ints and back again. This happens to work on machines where pointers and ints are the same size (11/70, 8080+Z80) but will it work on new 16bit machines? Will it keep working? Memories are getting cheaper, and already there are machines whose address space is bigger than the intuitive size of an int on that machine. 4) He #ifdefs himself to death, making the code very hard to read in sections. 5) He confuses int op[16] (an array of 16 ints) with int (*op[16])() (an array of 16 pointers to functions returning int). 6) His include syntax is different from standard C for no obvious reason. 7) From the code it would appear that #includes cannot be nested (I havent tried this) I now have the thing working, sort of, under UNIX. At least it stopped giving me core dumps if I stick to the straight and narrow and dont give it bad file names. It might be possible to make it work sensibly under UNIX with a lot more work. It isnt obvious that it would then compile itself any more. I am trying to isolate UNIX changes in ifdef UNIX sections, but it isnt always obvious what it will and wont compile itself. But, and this is a big but, it seems to have the 8080 architecture hard-wired into it, and ripping that out will be more work again. I must say, to its credit, that it is a great improvement over release 1 of the compiler, which was even more stuck in the 8080 grove. Fixing all these things up might be more work than writing a proper subset compiler. Does anyone else working on this compiler have any thoughts on the subject? Michael Wagner, UTCS