Brent Dax wrote:
Dan Sugalski:
# ... Later the # ancillary routine may be nonexistant if we build up the function # headers on the fly and embed the destination function into them.
Oh JITters... ;^)
jit/i386 has already code to call (specific) functions e.g. vtable funcs or pmc_new_noinit. The main differemce is, that the function signature (for vtables) comes from jit_info i.e. the definitions in core.ops.
The nci an is abstraction of this. *But*, what I really don't like in nci.c are the string functions. Can't we define, the signature is plain ascii?
Second: don't we need a signature for passing a Parrot_Interp* too?
If this is going to be the basic XS mechanism, why don't we just ask the user (or, more likely, the preprocessor) to write an argument-handling routine?
JIT will need it anyway. leo