RCpp containers lack `cbegin` / `cend` iterator pairs

[SylvainCorlay]

Methods cbegin and cend differ from begin and end in that they always return a const_iterator even when the object is not const.

[eddelbuettel]

We never had them in the API.

It's been a while since this came up, but I think part of the reason was than we could never guarantee const-ness of the underlying SEXP.

So what you write in xtensor-stack/xtensor-r#24 is somewhere between wrong and silly. Can't be our bug if we never had it. Your bug for assuming (without checking) we do.

But presume we can consider pull requests which implement / mocking / aliasing these.

[dcdillon]

First, Rcpp is c++98 compatible for CRAN reasons. So cbegin() and cend() don't even exist.

Second, there are legacy constness problems. So to create a compliant cbegin() and cend() across the board is a bigger issue.

Now I'm not sure there is a way to write c++ code that inherently depends on cbegin() and cend() even in newer versions. So I would stick with the given interfaces.

Other than that, you could tackle the constness problems, which should be pretty straightforward in the vector types for numbers. It's a touch trickier for character vectors, but certainly not insurmountable.

In summary, the interfaces are more or less c++98ish. They do need to be updated (as well as constness corrected), but this is not a barrier to usage. Constness would be a good project. Let's get it rolling!

[SylvainCorlay]

So what you write in xtensor-stack/xtensor-r#24 is somewhere between wrong and silly.

Wow... On a technical standpoint, xtensor assumes that the underlying shape type implements the standard methods of STL containers. Maybe that was wrong and silly.

Can't be our bug if we never had it.

My intent when opening an issue in the tracker for a project is not to throw blame on anyone but help improving the project as I am using it.

[eddelbuettel]

Absolutely, and we welcome that. But Rcpp interfaces R, and some things are simply different from what you may have seen for Python and other languages. We simply aim to clarify where your assumptions may have mislead you.

As @dcdillon says, this would be a good project to have, and not a trivial one to get right. Maye we should do some talking and sketching.

As for the standards methods of STL containers, we do have begin, end and much more. We also warn that certain access functions are there, yet should possibly not be used (because underlying storage is always SEXP based making some ops rather costly).

[SylvainCorlay]

First, Rcpp is c++98 compatible for CRAN reasons. So cbegin() and cend() don't even exist.

Second, there are legacy constness problems. So to create a compliant cbegin() and cend() across the board is a bigger issue.

I am aware of the novelty of cbegin / cend. Although adding them would not be a problem from a C++98 perspective.

Constness would be a good project. Let's get it rolling!

Yes, as I was posting the issue, I was looking at what it would take to make the Rcpp sequences be const-correct.

[dcdillon]

I'm happy to help. I've already solved the char vector problem. I just don't have time to be the point person for constness currently.

[dcdillon]

To @eddelbuettel point, there are also std algorithm problems...specifically with char vectors. If we could launch a project to fix this, that'd be good for everyone.

[dcdillon]

Iirc string_proxy is not move constructible. This breaks any std algorithm that reorders a container.

While this isn't necessary for numerical computation, we should just do it all at once and get it right imho.

[coatless]

@dcdillon: I'll throw some time behind this. Do you have a general game plan? Or should I reference the char vector work you've done?

[dcdillon]

It's not anywhere you can see. It lives in infamy on my local computer. If you can spearhead, I can help review and do the string stuff.

[dcdillon]

We have started work on this. It's in the preliminary stages, but I think we have things in a passable state for Vectors of primitives. The other issues with more complex types have not been addressed yet. The code is available at https://github.com/coatless/Rcpp in the feature/const-iter branch. Any help testing would be appreciated.

[coatless]

Direct code link: https://github.com/coatless/Rcpp/tree/feature/const-iter

[eddelbuettel]

This got merged a release ago and we forgot to close this.