16 comments

[ 2.7 ms ] story [ 24.0 ms ] thread
I just looked at the repo, haven't tried this yet. Could be very useful to me if tensorflow, pytorch etc. are callable
BTW, guenchi thanks for writing this code! Much appreciated.
I have been working hard to promote the construction of the (Chez)Scheme community for a year. Being useful to you is the best gift for me.

By the way, we have a scheme binding of tensorflow. That calling direct by tensorflow's C API.

FLI is in order to solve the problem of calling Numpy by Scheme which is lack of C - api.

I'm pretty neutral to Scheme and the other Lisps, but at the very least this seems like a cool project.
Racket has this for C-based APIs. My mind is kind of wondering how this would ever be stable/maintainable to use for Python, but Lua seems possible but I would rather do my scripting in Racket then Lua and use the C APIs directly.

https://docs.racket-lang.org/foreign/index.html

Looking at Julia (Which has a no "boiler plate" philosophy when using C or Fortan) the function is just called.

Interesting to look at for comparison

Julia ccall- https://docs.julialang.org/en/v0.6.1/manual/calling-c-and-fo...

Racket FFI - https://docs.racket-lang.org/foreign/index.html

I am researching julia's code.

https://github.com/JuliaPy/PyCall.jl/blob/master/src/PyCall....

I guess it is like this: Importing the pycall library is to turn on the python interpreter thread and not interrupt it. Then you can call any python library you want.

The GIL is mostly managed by CPython and the host application doesn't touch it directly. Host extension modules may use PyEval_SaveThread et al. just like regular extension modules to release the GIL during blocking operations when called by CPython. There is a separate API for fondling the GIL when you're in a host-created thread (not created by CPython) and want to call CPython. Those are orthogonal (called by CPython vs. calling CPython).
> My mind is kind of wondering how this would ever be stable/maintainable to use for Python

CPython has a stable C API for interacting with the interpreter and Python objects. Presumably FLI goes through this API to interoperate with Python, just like any C/C++ application embedding CPython would.

Edit: Yep, https://github.com/guenchi/FLI/blob/master/pycall.sc

Yes, it embeds the python interpreter into Scheme via C-api.
How does Chez Scheme compare vs. CHICKEN Scheme? For a layman, they both seem to compile to binary; Chicken via C, Chez maybe not? Which one has a bigget community?

Also, do you know of any static typing extension for Chez?

Chez is compile direct to Native Code

Chicken has much more library and bigger community.

BUT...

Chez is the only commercial compiler (before open source) and very stable and fast...

Chez is by far the fastest of all, similar to LuaJIT. Chicken has more libs.

Static? no, maybe Stalin works on Chez

Can anybody explain how this works ? Is it only works for interpreted languages ?
Every language with FFI has the same ABI. Because they use C Language's ABI to port another language.

Looking at it from another perspective, any language can call another language through this ABI.

So any language with FFI can call any compiled binary code. For interpreted languages, you must embed the runtime (or interpreter).