[libc-commits] [PATCH] D69421: [libc] Header generation scheme.

[Siva Chandra via Phabricator via libc-commits]

sivachandra added inline comments.


================
Comment at: libc/spec/stdc.td:5
+  let Functions = [
+    Function<"acos", "double", ["double"]>
+  ];
----------------
sivachandra wrote:
> sivachandra wrote:
> > theraven wrote:
> > > theraven wrote:
> > > > A minor issue, but it would be nice if the argument types were TableGen types and not strings, so that we can avoid accidental header pollution.  For example, there are a number of libc headers that define interfaces that use `size_t`, but that do not export `size_t`.  To make these work, implementations typically define `__size_t` in an internal header and then `typedef __size_t size_t` (guarded against multiple definition) in the headers that are expected by the standard to export it.
> > > Additionally, it would be nice to decouple attributes here.  For example, `restrict` and `_Nonnull` can be specified inline (though the GCC equivalent can't), but things like the printf-like attributes are properties of the arguments that need to be specified as function attributes.
> > > 
> > > In an ideal world, we'd also have SAL annotations, but these are currently lacking a clang implementation.
> > About "decouple attributes", Can you give some examples of what you are describing here? I probably understand vaguely, but I want to see an example to make sure.
> [Side note: We have decided, on another patch, that items like `size_t` come from freestanding headers and so we will make use of them instead of making LLVM-libc re-implement/re-define them.]
> 
> I understand and agree that it would be good to have TableGen types for arg types But, how deep/far do we go? For example, should there be a table gen type `size_t` and another one for `size_t *`?
> 
> What exact problems do you foresee with using strings instead of TableGen types? One obvious problem is typos. Are there any other?
This is not to say we should not do TableGen types. I have now uploaded a new diff with only TableGen types.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69421/new/

https://reviews.llvm.org/D69421