On Thu, 23 Nov 2006, Basile STARYNKEVITCH wrote:
For a kernel stuff, I tend to think that user interaction should be done thru a helper application. Hence !i still believe that something like lightning or libjit is the best approach. This helper application could statically check the high-level code input by the user (so you might write a type inference system for this) and translate it to something else (a low level representation) that the kernel would use to generate machine code.
That's true. But that would require designing a language, writing a parser, and basically a lot of work. I agree it's a more elegant, correct, and ideal approach. Right now I don't think anyone would notice if I just got C working -- I am about 95% there, and really this is just a small part of a much larger problem we are trying to solve.
And BTX tinycc generate machine code which is not better than what you'll get with lightning, and which could be worse that what you get with libjit.
Ok, point well taken.
Kernel code have severe constraints - you mentioned the 4k stack . So you probably should put entirely a full compiler in it. And as you mentioned, C is already too low level for you.
Yes. *Sigh*. Yes, I know this approach of using tcc is hackish as best. I'll cope with it. Ideally I wish I had more time to work on this and to do it correctly.
-Calin _______________________________________________ Tinycc-devel mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/tinycc-devel
