[llvm-commits] [cfe-commits] PATCH: Add VariadicFunction class template to LLVM's ADT

[David Blaikie]

Test cases (more as motivating examples than anything - I realize the
logic is fairly simple)?
Should this pass an llvm::ArrayRef rather than a pointer & size?
(they're implicitly constructible from arrays anyway - so it'll
require almost no change (dropping the "N" from the variadic macros
and default constructing in the zero-arg case - or perhaps even
passing no args in the zero-arg case, then users of this API could
choose whether they could accept no args by simplify using a default
value for the ArrayRef parameter, or not)

Where is this tool ultimately going to reside - inside clang itself?
or could we have it build separately, and thus use C++11 features
directly in its implementation?

- David

On Thu, Dec 15, 2011 at 3:55 AM, Chandler Carruth <chandlerc at google.com> wrote:
> Hello folks,
>
> This patch adds a generic collection of class templates to the LLVM ADT for
> building variadic-like functions in C++98. The idea, from the comments, is:
>
> /// Suppose we need a variadic function like this:
> ///
> ///   ResultT Foo(const ArgT &A_0, const ArgT &A_1, ..., const ArgT &A_N);
> ///
> /// Instead of many overloads of Foo(), we only need to define a helper
> /// function that takes an array of arguments:
> ///
> ///   ResultT FooImpl(const ArgT *const Args[], int Count) {
> ///     // 'Count' is the number of values in the array; 'Args[i]' is a
> pointer
> ///     // to the i-th argument passed to Foo().  Therefore, write
> '*Args[i]'
> ///     // to access the i-th argument.
> ///     ...
> ///   }
> ///
> /// and then define Foo() like this:
> ///
> ///   const VariadicFunction<ResultT, ArgT, FooImpl> Foo;
>
> We used this pattern extensively to build up the AST matcher library on top
> of Clang (and so I've included cfe-commits in case anyone there is
> interested). In preparation for mailing patches for those matchers, I'm
> sending a patch that just adds the generic utility to the ADT library as
> that seems like a natural home for it.
>
> Previous versions of this bit of helper code used a fully expanded source
> file that would be hard to maintain. I was worried about using the
> preprocessor to expand these constructs, but it turned out *much* easier
> than I had feared, and I think the result is much easier to read and
> maintain. That said, if anyone dislikes the preprocessor hammering, I'm
> happy to switch to a more manually expanded form.
>
> I've also included Doug so that he can rant about adding hacks to LLVM/Clang
> to work around the absence of variadic templates. ;]
>
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>