Help understanding why libfive_tree_render_mesh would return nullptr about libfive HOT 7 CLOSED

Oh my god, it worked with __stdcall. I'm out. Thanks for the booth 🙂

from libfive.

For what it's worth, this other test segment passes:

//home/lee/Code/lee/libfivejs/index.js:234.1,251.2
{
  let str = "libfive_tree_print ";
  const a = x();
  const b = y();
  const c = z();
  const a2 = a.square();
  const b2 = b.square();
  const c2 = c.square();
  const r_ = a2.add(b2);
  const r = r_.add(c2);
  const d = r.sub(1.0);
  const ptr = d.toString();

  if (ptr !== "(- (+ (square x) (square y) (square z)) 1)") throw Fail;

  str += "passed";
  console.log(str);
}

(Same test, just more idiomatic)

//home/lee/Code/lee/libfivejs/index.js:258.2,253.1
// Again, but more golfy :)
{
  const [a,b,c] = [x(),y(),z()].map((d) => d.square());
  const r = a.add(b).add(c).sub(1.0);
  if (r.toString() !== "(- (+ (square x) (square y) (square z)) 1)") throw Fail;
}

from libfive.

libfive_tree_save_mesh also works... properly outputs a cube. Hm.

from libfive.

Does it maybe have something to do with new libfive_mesh ? Is this equivalent to malloc'ing?

from libfive.

I'm thinking that everything is working fine, and it's the FFI (https://koffi.dev) which is having problems, potentially with what I mentioned...

Edit: Could it do with the call type you think? i.e. fastcall, stdcall, etc

Edit: the more I research, the more I think that maybe using new from C++ is a bit of problem when providing the C api, although logically I don't understand why. It should be fine.

Edit: nope, changing it to malloc didnt do the trick.A

Edit: Yeah ok, adding an "output" argument onto render_mesh makes it work. I think the issue is this FFI system doesn't support using pointers as return types, unless they're opaque (in which case you don't access them) or strings (which it handles just fine).... gah, that really sucks, because everything else works!

Edit: So https://github.com/Koromix/rygel/blob/dd1cefe9c8e4efbff7c8b2ab71fcde21a0e81323/src/koffi/src/call.cc#L953 must be the reason. Anything allocated outside of the nodejs context is marked "External", otherwise, using output parameters, it's possible to have much more type information... Not sure how to fix it though.

Edit: from what I understand, it makes the most sense is beneficial for these systems to allocate and deallocate memory and pass these objects to the API. Then the GC can take care of things

from libfive.

Welp, upon revisiting this morning, it appears I was just very tired, and what I saw was just my old custom change to libfive working... So back to being screwed :). At least I can do mostly everything but this is necessary if I want to have a somewhat performant rendering loop (I'm going to use node-sdl with it). Is it possible to provide alternatives that take in output parameter pointers?

from libfive.

Koromix/koffi#35

from libfive.