Do not use dir.exists to retain backwards-compatibility

[epipping]

The file R/Attributes.R uses the function dir.exists() which
was only added in R 3.2.0. The relevant line was added to
R/Attributes.R between 0.12.5 and 0.12.6.

[eddelbuettel]

Good catch.

[epipping]

If compatibility with versions older than 3.2 is important (debian 8.8 still only has 3.1) one might want to turn the dir.exists back into file.exists instead.

[eddelbuettel]

Debian 8.8 is irrelevant as its sources for Rcpp match its sources for base R. Plus, we have the actively supported backport for R available via CRAN.

We could use the backports package, or at least Suggests: it if we cared.

Team: Anybody see a problem depending on the now-two-year old R 3.2.0 ?

[kevinushey]

I'm guessing there are a lot of users in enterprise environments who are stuck with R 3.1.x or even R 3.0.x who would appreciate if we could remain compatible with R 3.0.0 and above, so I would be marginally in favor of just using file.exists() or utils::file_test("-d", ...) to be backwards compatible -- especially since the change is so small in this case.

[eddelbuettel]

Yes, that is a good policy. I can look into that while traveling the next two days. Of course I would be equally happy to receive a short PR :)

[epipping]

utils::file_test("-d", ...) is the perfect solution for this problem. Since I see no disadvantages in following that route, for simplicity, I've now hijacked my own pull request and turned it into this different solution for the same problem by force-pushing.

[jjallaire]

Yes, we have to remember that we bump our required version we implicitly bump the required version of 92% of CRAN (via recursive dependencies).