Return null start for empty vectors

[Enchufa2]

Closes #1461. This happens because this returns 0x1 for empty vectors. The question is whether we should do the same check in here. I think we should.

Checklist

• Code compiles correctly
• R CMD check still passes all tests
• Preferably, new tests were added which fail without the change
• Document the changes by file in ChangeLog

[eddelbuettel]

That looks great. I could turn on the big bad rev.dep machine but I am not even sure it's needed.

[eddelbuettel]

This looks good -- test on entry is the way to go. There might as always be other code paths but this is at a minimum a start if not even a solid 'squash' of that bug.

[kevinushey]

Are we sure std::copy() is going to accept a null pointer here? If we don't already have it, a unit test would be nice -- presumably ASAN / UBSAN builds of Rcpp with those tests would catch this issue.

I'm also fairly certain the actual issue here comes from this bit of cruel code.

https://github.com/wch/r-source/blob/6b5f72eebd9211766bd159e017a167385c93772e/src/include/Rinlinedfuns.h#L135-L142

[kevinushey]

Perhaps we can return dataptr(x) instead of NULL? With the appropriate casting.

Or, alternatively -- instead of using __ACCESSOR__ here, just use DATAPTR_RO, and cast to the appropriate type. But we're sort of cheating since DATAPTR_RO is supposed to only give read access to memory, and these might be writable...

I just hate the idea of this zero-length check becoming necessary. :|