> What libraries or tooling would you like to see written for Common Lisp next?
I would like the entire Python scientific ecosystem be ported to Common Lisp instead of Python. Lisp is vastly superior for numerical work.
What would be required for this to happen?
I think empirically we can say that people who wrap Intel MKL, BLAS, etc. prefer a stable C-API with stable C functions that access the internals and do not want an FFI.
The native FFI from SBCL is difficult to handle and performance is somewhat unpredictable due to garbage collection.
I know this is a hard problem that alternative Python implementations like PyPy also fight with, but if SBCL somehow managed to publish a C-API that is as fast as CPython's (single threaded is good enough for the start) and retain the excellent speed for pure Lisp, I think it might be a great success.
Even tricks like automatically activating some form of a GIL when loading a C module would be completely acceptable. In general, roughly copy the Python C-API, it is safe to say by now that people want this.
smcn
Favourite time of the year!
I am, however, continually surprised that SLY doesn’t have more mindshare. Similarly jonathan. I guess cl-json and SLIME are _good enough_ for most people.
I would like the entire Python scientific ecosystem be ported to Common Lisp instead of Python. Lisp is vastly superior for numerical work.
What would be required for this to happen?
I think empirically we can say that people who wrap Intel MKL, BLAS, etc. prefer a stable C-API with stable C functions that access the internals and do not want an FFI.
The native FFI from SBCL is difficult to handle and performance is somewhat unpredictable due to garbage collection.
I know this is a hard problem that alternative Python implementations like PyPy also fight with, but if SBCL somehow managed to publish a C-API that is as fast as CPython's (single threaded is good enough for the start) and retain the excellent speed for pure Lisp, I think it might be a great success.
Even tricks like automatically activating some form of a GIL when loading a C module would be completely acceptable. In general, roughly copy the Python C-API, it is safe to say by now that people want this.