[CM] On FFI bindings

[Woody Douglass]

Returning an environment with bindings from init-func sounds very elegant; essentially that's what i'm doing anyway. I'm all for this as long as it doesn't prohibit adding symbols to the root environment too



On 2020-10-22 20:16:53-04:00 cmdist-bounces at ccrma.Stanford.EDU wrote:

Another thought on the semantics:
- keep init_func behavior as it is (legacy, snake_case, backward compatibility)
- passing init-func (kebab-case) new behavior: calling with the entrypoint in the same manner as
  the FFI functions: s7_scheme* sc , s7_pointer args. args being (the-passed-env 'args) : being either
  something or #f
Wonder if there's a point in having the called entry point also return s7_pointer. It would be
elegant though as far as consistency goes: calling any s7 FFI function as the init-func.
Possible use: init-func returning an environment with bindings.

On Thu, 22 Oct 2020 at 22:18, Christos Vagias <chris.actondev at gmail.com<mailto:chris.actondev at gmail.com>> wrote:
If I understood correctly you get a segfault when having the C function with 2 args (s7_scheme and s7_pointer)
but passing only s7_scheme.
This seems relevant: https://stackoverflow.com/questions/12572575/i-can-call-a-function-imported-with-dlsym-with-a-wrong-signature-why
and particularly the answer saying
"C uses cdecl call conversion (so caller clears the stack) [..]
But actually behavior is undefined"
In any case, I'm really noob in this area so that's all I can contribute.

So my 2 cents:
- when not having passed init_args: calling the C init_func(s7_scheme)
  (serving also as backwards compatibility)
- when init_args is present, calling init_func(s7_scheme, s7_pointer)

On Thu, 22 Oct 2020 at 21:57, <bil at ccrma.stanford.edu<mailto:bil at ccrma.stanford.edu>> wrote:
> how does C behave if you assume that the called function has a
> signature
> of (s7_scheme* sc) and you try to pass (s7_scheme* sc, s7_pointer args)
> ?

I tried it with the tlib example, and if you declare args in C, but
don't pass them in scheme (i.e. no init_args in the environment),
I get a segfault.  The other way (no args declared, but you pass and
use them anyway) seems to work -- strange!  This is in gcc 10.2
in Linux.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://cm-mail.stanford.edu/pipermail/cmdist/attachments/20201023/bda05622/attachment-0001.html>