Bindings generation from C header files #91
Conversation
Numba's API wrapper is no longer used.
|
Fix #95 |
|
Great work so far. See how zig does it: https://ziglang.org/learn/overview/#integration-with-c-libraries-without-ffibindings. Namespacing would be useful. If we can do that, we can ditch explicit naming of all the functions, typedefs etc. |
| (c-lisp | ||
| ,@(include | ||
| (shape.h) | ||
| () |
There was a problem hiding this comment.
I don't like this syntax that much because of empty lists and non named stuff
There was a problem hiding this comment.
Right. I'm open to writing a new macro that wraps this macro, takes no arguments and includes all available functions + structs they depend on.
Available functions can be scraped from the JSON AST.
There was a problem hiding this comment.
Although the ability to have unspecified arguments is an important capability. So maybe we could implement keyword arguments in Prelisp?
Perhaps something like this, in sexp:
,(macro-fn parg1 (parg2_1 parg2_2) ((karg1 . val1) (karg2 . val2)))and equivalent Python call would be
macro_fn("parg1", ["parg2_1" "parg2-2"], karg1="val1", karg2="val2")There was a problem hiding this comment.
Let's look up the syntax for lisp for named arguments and copy it!
There was a problem hiding this comment.
Alright, Guile supports keyword arguments, so we can use that.
; Procedure definition
(define* (func #:key arg)
arg)
; Procedure call
(display (func #:arg 'value))result:
value
Although there are a lot of Scheme implementations (including Guile) that support some form of keyword arguments, the Scheme spec doesn't provide keyword arguments.
There was a problem hiding this comment.
The only issue here is that guile-json doesn't seem to be able to handle it. When the following call is passed through sexp-json.scm,
(define ((main int))
,(func 'pos-arg-val #:kw-arg-val)
(ret 0))the following error is thrown:
json/builder.scm:189:9: In procedure json-valid?:
Throw to key `json-invalid' with args `(#:kw-arg-val)'.
So we'll have to write our own logic in sexp-json.scm to handle keyword arguments in macro calls.
There was a problem hiding this comment.
try reading the this using (read) first and then see how it works. It's possible it's just a symbol?
| ,(error_check (call cuDeviceGetCount (ptr-to devCount))) | ||
| ,(error_check (call cuDeviceGet (ptr-to device) 0)) | ||
| ,(error_check (call cuCtxCreate (ptr-to context) 0 device)) | ||
| ,(error_check (call cuCtxCreate_v2 (ptr-to context) 0 device)) |
There was a problem hiding this comment.
cuCtxCreate happens to be a macro. And macros won't show up in the JSON AST :/
So with the current framework it's not possible to infer what cuCtxCreate is supposed to be
There was a problem hiding this comment.
Okay, the issue is a little more complicated than that. It's defined both as a macro and as a function:
// ...
#define cuCtxCreate cuCtxCreate_v2
// ...
CUresult CUDAAPI cuCtxCreate(CUcontext *pctx, unsigned int flags, CUdevice dev);
// ...As far as I can see, neither definition is within a conditional (e.g. #if, #ifdef) block.
There was a problem hiding this comment.
Look up zig and see how it parses it: https://ziglang.org/learn/overview/#integration-with-c-libraries-without-ffibindings
This commit introduces the following changes: * Keyword arguments can now be specified in macro calls like this: `,(macro_func (#:kwarg val))` This is equivalent to the following Python call: `macro_func (kwarg=val)` * The following tests: + parser/keyword.sexp -- Handling JSON serialization of keywords + prelisp/kwargs.sexp -- Passing keyword arguments to macros + prelisp/errors/invalid-keyword.sexp -- Error test
2 participants
This is an effort to automate the process of generating bindings to libraries using their C header files. Parsing and resolving types is offloaded to Clang/LLVM, while
gen_bindings.pymaps LLVM types to C-Lisp types and exposes the binding generation machinery via C-Lisp macros.