Upstreaming `std::tuple<>` support about rcpp

[jonclayden]

RcppArray v0.3.0, with the tuple support added, is now on CRAN.

[eddelbuettel]

Howdy and thanks for suggesting that. We will take a look. It may otherwise be easiest to do this in a small add-on package as e.g. was done in RcppArray (also supporting std::span if C++20 is used).

[jonclayden]

Very happy to look into merging this into RcppArray, as it does seem like a natural extension and saves creating another small extension package for a similar purpose.

[eddelbuettel]

Sounds good to me, @andrjohns is closer to the metal for a concrete use case. I am sure we can cook up a unit test or two, I always liked how you set up tester packages in RcppArray (and Rcpp has long done similar stuff for parts of its tests).

[eddelbuettel]

Gee I just realized in horror that I think I never experimented / reached out to @jonclayden here! John, any 'feels' about extending RcppArray with this (pro: it seems to fit; con: it may clutter the package) or whether this should be something on its own (pro: self-contained is good ?; con: clutters package system?).

[jonclayden]

The commit referenced above provides a version of the implementation that avoids using helpers from C++14 (index_sequence) and C++17 (apply()), to try and maximise compatibility. Instead it uses a nested helper struct holding references to the tuple and the R list, each "level" of which is responsible for an element of the tuple. I have tests for the as() direction (though not yet for wrap()) and it seems to work. I don't imagine this will be noticeably less efficient, and it compiles with CXX_STD=CXX11, but testing/feedback is very welcome.

[jonclayden]

I've added a tuple client stub package now, and unfortunately wrap() chokes on this assertion, whether my implementation or the version above using apply() is used. The specialisation is declared before including the full Rcpp.h but somehow dispatch isn't working out correctly. I'll need to have another look with a clearer head...

[jonclayden]

The problem seems to have been including the wrap() implementation within the Rcpp::traits namespace, when it should just be in Rcpp. The Rcpp-extending vignette threw me here, I think, because it seems to indicate that partial specialisations should live in Rcpp::traits (section 2.3), although that does seem surprising – am I misreading it? Anyway, that seems to have been the mistake relative to the version above, now corrected.

@andrjohns I'll add you as a contributor to the RcppArray package, assuming you're happy with that. If the current implementation works for you then I'll put together a release for CRAN.

[eddelbuettel]

Yes, that documentation / implementation could do with a rewrite. It also throws me for loops at times. So twice the thanks for working through it !!

[eddelbuettel]

We can close this as RcppArray took care of it. Three cheers for that!