adds empty() to [const_]string_proxy for more efficient checks for empty strings

[jmarshallnz]

Timings from R:

f1 <- cppFunction("LogicalVector is_empty(CharacterVector v) {
                     const int n = v.size();
                     LogicalVector e(n);
                     for (int i = 0; i < n; i++)
                       e[i] = v[i].size() == 0;
                     return e;
                  }")

f2 <- cppFunction("LogicalVector is_empty(CharacterVector v) {
                     const int n = v.size();
                     LogicalVector e(n);
                     for (int i = 0; i < n; i++)
                       e[i] = v[i].empty();
                     return e;
                  }")

chars <- rep(paste(rep("x", 100), collapse=""), 10000000)
chars[sample(10000000, 100000)] <- ""

> system.time(f1(chars))
   user  system elapsed 
  0.176   0.000   0.177 
> system.time(f2(chars))
   user  system elapsed 
  0.079   0.001   0.080 

Times improve rapidly as the string length gets longer, saving going via a string.

[eddelbuettel]

As you can tell from the 'Failed' above, it does not pass the regression tests. As such we are unlikely to adopt it. Could you please try and and see why it fails?

[thirdwing]

It seems a problem in the ppa (edd/misc). The test never started.

https://travis-ci.org/RcppCore/Rcpp/builds/56843011#L712

[eddelbuettel]

Oh, sorry, am on vacation and with less-than-usual computer access. Failed tests / timed-out tests can just be restarted, which I just did. Sadly they now idle for a while before starting.

I may have to invoke apt-get with options directly rather than relying on travis-tool.sh if something changed, Either way I am on it now -- thanks for the heads-up.

[eddelbuettel]

Looks like it worked. I will have to reconsider the pull request :)

[jmarshallnz]

No worries - just a nice to have in some cases prior to taking the hit of going to std::string or similar for further processing.

[eddelbuettel]

Learned something new in this issue ticket for Travis that I was cc'ed on: one now needs to declare a dependency on 'sudo' on new repos. The PR was a 'new repo' at least for @jmarshallnz and so this failed, but didn't when I re-ran it because my repo was pre-existing. Oh the joy :) At least now we have an explanation.