I'm a bit worried about extending to meet allcustom objects defined.Matrix is one thing as it part of the recommended packages. Not so withslam.

Also, please modify this commit to respect spacing.

I think we managed to useslam,SparseM, andSciPy asoptional pieces in the tests. That way they do not bubble up into Depends. I would much prefer to keep it that way if we could.

@coatless

• I don't think that taking into account slam format adds significant charge on the package. The code is small and compared to previous executions only one type check is added;
• you are right that Matrix is in recommended packages and not slam, but Matrix is much slower to load than slam. So if someone manages to get all he wants from a small package, why to bother him to force to drag a big one?
• white spaces are respected now.
  @eddelbuettel I didn't find any mention of slam in unit tests (a part in comments) so I've just put the test inif (suppressMessage(require(slam))) {} block.

Correct. They way you have the test conditioned we should only need a Suggests:. And on the C++ side it just adds another type. I see no harm in that.

But as a general rule (to both (!!) of you): I generally still prefer issue ticket discussion so that everybody can pipe inbefore filing PRs. But you two of course have built upsome trust by now, but I remain a very mean maintainer (even if I just merged@coatless 's PR to Rcpp....)

@eddelbuettel in my defense, I didn't think you would appreciate an issue ticket for a six character change on a single line followed immediately by a PR ticket.

To make it plain, we half a dozen communication channels between us. Pick one. Any one.

Randomly submitting PRs is always the worse alternative.

I realize that even with this patch we still need to load Matrix package to get it working. This invalidates my comment about time differences in loading Matrix and slam :(

Oh well. It'll be in 0.8.300.1.0 which I am currently rev.dep testing. Some fixes by Conrad in 8.300.1.