News for hackers. Stuff that matters.
Re:Type-erased generic functions for C: A modest non-proposal by rwmj on Thursday October 03, 2024 @03:07AM (#41728016)
Re:Type-erased generic functions for C: A modest non-proposal by eqvinox on Thursday October 03, 2024 @04:50AM (#41728664)
Re:Type-erased generic functions for C: A modest non-proposal by naasking on Thursday October 03, 2024 @09:45AM (#41730756)
That monomorphizes every definition, and the point of this proposal was to avoid that.
Re:Type-erased generic functions for C: A modest non-proposal by rwmj on Thursday October 03, 2024 @10:48AM (#41731312)
That's true, although I wonder if a "sufficiently smart compiler" could combine compatible definitions. Also note it's in a header file and most of these functions are minimal enough they would be inlined.
Re:Type-erased generic functions for C: A modest non-proposal by eqvinox on Thursday October 03, 2024 @11:31AM (#41731691)
I'd distinguish here and not label one-liner "cast wrappers" as monomorphization. And then you can in fact do a lot with macros.

(My definition of a "cast wrapper" is: function body contains exactly one function call, mostly forwarding arguments. Pointer offsets¹ on arguments/retval are allowed, as well as adding sizeof/offsetof/constant integer arguments, but nothing else.)

¹ this may include some conditionals to keep NULL as NULL when applying a pointer offset.

I would expect the compiler to be able to optimize these away in 99% of cases, but TBH I don't care if it doesn't and I'm littering a bunch of small wrapper functions.

Re:Type-erased generic functions for C: A modest non-proposal by variadix on Thursday October 03, 2024 @02:01PM (#41733142)
You can get better properties by using the include itself to do the macro expansion, e.g.

#define VECTOR_TYPE int #include “vector.h”

This will let you step into the macro generated functions in GDB and doesn’t require putting the function creation in macros (no backslashes, can write the implementations like normal code)

You can also get syntax like this that way Vector_Push(int)(&vec, 1);

Which imo does a better job distinguishing the type from the name

Re:Type-erased generic functions for C: A modest non-proposal by david2ndaccount on Thursday October 03, 2024 @05:29PM (#41735215)
I wrote a blogpost describing this technique. https://www.davidpriver.com/ctemplates.html
Re:Type-erased generic functions for C: A modest non-proposal by eqvinox on Friday October 04, 2024 @01:27AM (#41738120)
I agree that the debugging and maintenance experience is much better that way, but it comes at the cost of very poor "DX". We, discussing this here on this thread, understand what's going on. But sadly at least for the projects I'm involved in, the average developer does not.

TBH, rather than fancy high-meta generic support, I'd be much more appreciative of ISO C adding something like "_Include" to the preprocessor, e.g.

  #define DEFINE_MY_WRAPPER(foo, bar) _Include("wrapper.h")
  /* "foo", "bar" remain accessible in wrapper.h */
I tried implementing this in GCC but didn't get very far, it runs a bit counter to how includes are handled currently.

P.S.: I ended up doing this https://github.com/FRRouting/frr/pull/15876/files#diff-90783... instead; a tool that can be invoked to expand some DEFINE_FOO if needed for debugging. It's not used particularly often, though.