06:19 <proteusguy> Anyone seen a forth implemented in C that managed to get around the C calling convention and implement something more akin to next with JMP?

06:21 <remexre> unless you count implementing a VM, I don't think you can in standard C

06:22 <proteusguy> You would have to create another calling convention or subvert C's.

06:22 <proteusguy> some inline asm perhaps.

06:23 <remexre> perhaps; you might need "naked functions" too?

06:24 <remexre> not sure how much of a win that is over just using assembly at that point though; I suppose you get somewhat more portability

06:25 <remexre> but I don't know if you can convince the register allocator to not spill e.g. the forth stack ptr

06:25 <proteusguy> Didn't know what naked functions were before now... interesting idea.

06:27 <remexre> yeah, though note gcc documents: While using extended asm or a mixture of basic asm and C code may appear to work, they cannot be depended upon to work reliably and are not supported.

06:28 <proteusguy> I'm using clang and it appears to only allow inline asm inside a naked function... unfortunately.

06:28 <remexre> oh, yeah, clang in particular is gonna fight you hard about this sorta stuff

06:29 <remexre> iirc it even tries to translate computed gotos to a normal if-elseif-elseif-else chain

06:42 <proteusguy> goodness. :-) too bad you can't just goto a function rather than strictly a local label.

06:44 <remexre> yeah, the "C is fast" perception is mainly based on the super-aggressive optimizations that directly combat the "C maps closely to your hardware, and will let you do anything your hardware can do"

06:45 <remexre> on the other hand, do you know if tail-call elimination would work for your purposes?

06:45 <proteusguy> remexre, it does for a lot of it and I am doing that. but I was trying to make it, in general, as close as possible to what I'd implement in ASM, especially for the inner interpreter.

06:47 <remexre> fair; I'd imagine you could just #define NEXT return (++ip)(stack_ptr, rstack_ptr, ip, ...); though

06:48 <proteusguy> that's been another thought. I've got to see what the assembly output of the compiler looks like and see how it goes.

06:48 <remexre> yeah; if you're not familiar with it, godbolt.org is really nice for that

06:49 <proteusguy> remexre, ah yes I have seen this. Haven't used it as I'm pretty comfortable with the -s output on my command line but does look quite handy for testing something that doesn't have many dependencies.

06:54 <proteusguy> er and it doesn't compile my code. insists that there's no such thing as std::endian even when I have the correct include for clang10. strange. compiles perfectly on my own platform.

06:55 <remexre> huh, you might need to pass -std=c++20 or something?

06:55 <proteusguy> Ah likely so.

06:56 <proteusguy> yep that did it.

16:08 <MrMobius> proteusguy, what about function pointers?

16:23 <proteusguy> MrMobius, calling them still invokes the C calling convention. If you JMP to those addresses things have to be setup as the function expects or it'll all crash.