[R-SIG-Mac] OSX Binary Installation and testInstalledPackages()

[Marc Schwartz]

Simon,

Thanks so much for making this change!

I downloaded the latest nightly and ran all tests (testInstalledBasic("both") and then testInstalledPackages() for examples, tests and vignettes for both base and recommended) locally here, which all pass. This is great!

Will this change also show up in the next stable release on CRAN?

For those who may still be following this thread, be sure that you run these tests in an R session using 'R --vanilla' from the command line (OSX Terminal app in Applications/Utilities), rather than using R.app or other GUIs (eg. RStudio, etc.). You may have modifications in your .Rprofile or elsewhere that may conflict with these tests and the GUI's may similarly alter the default environment. In addition, you should run these tests before installing any CRAN or other third party packages, which may also result in some conflicts with these tests. 

Simon, thanks again for this change and the incredibly fast turn around!

Best regards,

Marc


On Feb 2, 2012, at 9:40 AM, Simon Urbanek wrote:

> The current builds now contains tests. As Brian noted size is probably not an issue anymore so it was the easiest to simply run install-tests in the build.
> 
> Cheers,
> Simon
> 
> 
> On Feb 1, 2012, at 2:18 PM, Marc Schwartz wrote:
> 
>> Hi Simon,
>> 
>> On Feb 1, 2012, at 12:08 PM, Simon Urbanek wrote:
>> 
>>> Marc,
>>> 
>>> On Feb 1, 2012, at 12:13 PM, Marc Schwartz wrote:
>>> 
>>>> Hi all,
>>>> 
>>>> I have been building R from source for a number of years on Linux and for the past 3 years, on OSX. Since circa version 2.9.0 I believe, there have been functions available in the 'tools' package to run post-installation tests of the base and recommended packages. These parallel the post-build from source 'make check-all' functionality, but can be run within R. They run the help file examples, more extensive package specific tests as well as vignette code when present.
>>>> 
>>>> When building from source, one can run 'make install-tests' to install the required files to run the more extensive package tests after the initial build, rather then just the examples in the help files.
>>>> 
>>>> On Windows, there is an option during the GUI installation of the pre-built binary, to install these additional test files.
>>>> 
>>>> When installing the pre-built binary for OSX, just to consider that path, I did not see an option to install these test files and I don't see any CLI functionality to replicate the installation of the test files that would be performed when building from source. A search of the R.framework tree did not yield any indication that the files are present, which would normally be in a 'tests' folder (eg. see https://svn.r-project.org/R/branches/R-2-14-branch/tests/).
>>>> 
>>>> So, are these files not installed or made available when using the OSX binary or am I missing something?
>>> 
>>> Yes, they are not available in the binary distribution. You can still use make install-tests when compiling your own R, and since they are arch-independent you can pick any architecture for that.
>>> 
>>> It would be certainly possible to include the test files in the distribution, but so far you are the first person asking about that, so it doesn't seem prudent. However, it would be even easier to just add a separate tar ball of tests to the http://R.reasearch.att.com site for nightly builds if that would satisfy your needs.
>>> 
>>> Cheers,
>>> Simon
>> 
>> 
>> Thanks for your reply and confirmation. I can certainly continue, as I have done, to build R from source. In my case, I build 32 bit R, as I don't have a need for the additional memory address space provided by 64 bits, even though I have 8Gb on my MBP. It takes about 20 minutes to build and another 40 to run the full suite of tests.
>> 
>> I was just doing some forward looking, considering alternatives to building from source and instead using the OSX binaries, both the stable release on CRAN and the patched versions that you kindly provide at the URL you list above. There are benefits to using the "official" binaries of course.
>> 
>> Providing a tarball of the tests folder, matching svn revs, for the nightlies, would certainly be helpful. Presumably there would need to be a brief README file so that folks would know how and where to install them. I can assist with that, if you desire. The default location would be /Library/Frameworks/R.framework/Resource/tests (eg. R.home("tests")).
>> 
>> My broad perspective is that these tests are in effect, an integral part of validating an R installation, which is something that I do. While this has special meaning for folks involved in clinical trials (eg. see 6.3 in http://www.r-project.org/doc/R-FDA.pdf), it seems logical that this would have value to folks in other domains as well. It would make it easier for useRs to fully utilize the functions in 'tools' that R Core has taken the time and consideration to provide.
>> 
>> Presumably, somewhere along the way around the time of 2.9.0, a decision was made to enable the test files to be optionally installed when using the Windows binary for ease of access, to enable that user base to be able to run the more extensive tests available. It would seem logical from a consistency perspective, to enable the OSX user base to have similar ease of access. I can't speak to what the various Linux packagers of R do, since that may vary across distributions due to differing packaging standards and it is possible that some may not even be aware of the tests functionality.
>> 
>> Thus, if and when convenient, it would seem ideal to make these a standard part of the OSX binaries. The entire tests tree is only about 5 Mb, so it does not materially add to local data storage requirements.
>> 
>> Thanks again Simon and let me know what I can do to assist. Feel free to contact me offlist.
>> 
>> Regards,
>> 
>> Marc
>> 
>> 
>>> 
>>>> I did not see anything relevant in the OSX FAQ for R and a search of the manuals and the archives yielded no joy. Presumably, one can just copy the files from the SVN repo above, being sure to match rev numbers during the check-out.
>>>> 
>>>> Thanks for any insights.
>>>> 
>>>> Regards,
>>>> 
>>>> Marc Schwartz