support for r 3.0.0

[javierluraschi]

Add support for r 3.0.0 to allow upstream packages like tibble to support 3.0.0 as well.

[hadley]

This also needs a comment to mark the source and license. And you'll need to add R-core to Authors@R in the DESCRIPTION

[hadley]

Can you please also modify the travis file to enforce this? (Ask @jimhester for details)

[jimhester]

The earliest travis version available on travis is 3.0.3, however there was a devtools bug in the CRAN version that prevents installation of packages with 3.0.3 (https://travis-ci.org/hadley/devtools/jobs/172912550#L1115). You will need to install devel devtools manually by adding something like the following to .travis.yml (untested).

r:
 - 3.0.3
 - oldrel
 - release
 - devel

before_install:
  - if [[ "$TRAVIS_R_VERSION_STRING" = '3.0.3' ]]; then git clone https://github.com/hadley/devtools ~/devtools && pushd ~/devtools && git fetch origin pull/1388/head:3.0.3 && git checkout 3.0.3 && R CMD build . && R CMD INSTALL devtools*tar.gz && popd;fi

install:
  - R -e 'devtools::install_deps(dep = T)'

[javierluraschi]

@hadley some tests are failing with the port from the original shallod_duplicate function. Instead, seems we should look at porting a more recent version: https://github.com/wch/r-source/blob/af7f52f70101960861e5d995d3a4bec010bc89e6/src/main/duplicate.c

Now... it looks like shallow_duplicate and duplicate are now sharing code through duplicate1. Being that the case, we could replace the call to shallow_duplicate with duplicate. However, I don't fully understand what would be the side effects of using duplicate in older versions of R that do not provide a shallow copy.

Would this be a performance downgrade? Or would this break lazyeval.

Happy to keep this PR open but it definitely needs review from more experienced folks familiar in the r core codebase and lazyeval.

[codecov-io]

Current coverage is 97.27% (diff: 100%)

Merging #87 into master will decrease coverage by 1.02%

@@             master        #87   diff @@
==========================================
  Files            15         15          
  Lines           354        367    +13   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits            348        357     +9   
- Misses            6         10     +4   
  Partials          0          0          

Powered by Codecov. Last update c155c3d...c90aa66

[hadley]

It's probably a minor performance regression, but it seems like it's going to be challenging to make lazyeval backward compatible with R 3.0.0.

[krlmlr]

Can we use Rf_shallow_duplicate() if it is available? In this case any potential regressions are limited to R 3.0 users. We should then also check if the R version lazyeval was built with a different R version in .onLoad().