I am creating my first package, which shall be installed through Github. I thought that Rtools was needed only for the person creating it. However, people that tried to install it using Github were asked to update Rtools. Is this really necessary?
Doing some research, I found this: https://community.rstudio.com/t/missing-rtools-should-i-be-worried/27817
One of the answers says the following:
"This means that if you are going to install packages that need
compilation, you also have to install Rtools in your system. "
This is the repo with the package: https://github.com/datazoompuc/PNAD_Covid/tree/master/R/datazoom_pnad_covid
What does this actually mean? How do I know that my package needs compilation?
I thought that Rtools was needed only for the person creating it.
Yes, if and only if you distribute it as a binary. Then the creator uses Rtools to compile and link, and the user just installs, and enjoys.
That is how CRAN works as CRAN compiles for Windows users.
GitHub, however, is foremost a source repository so the installation from GitHub is using a source mode ... and every user will need to compile, and hence have Rtools. (Unless the package and all its depedencies are R-code only.)
You can also have a package repository on GitHub using e.g. the drat package to create it, but that is getting us a little further from the original question.
Your package "needs compilation" — i.e. needs Rtools to install from source (on Windows) — if it contains C or Fortran components, i.e. if you have anything in the src/ directory of your package ...
If you, the package author, don't know if you have C or Fortran code as part of your package, then you almost certainly don't.
It's quite possible that devtools is being overzealous, i.e. detecting that users have a not-most-current Rtools and suggesting (requiring??) that they update it, even though it's not needed for this installation.
Related
I want to use the R package BOLTSSIRR available on GitHub in my R package, which I want to upload to CRAN.
I listed BOLTSSIRR under Suggests: in the DESCRIPTION file and made the link to GitHub available using Additional_repositories: https://github.com/daviddaigithub/BOLTSSIRR.
However, running R CMD check --as-cran I get:
Suggests or Enhances not in mainstream repositories:
BOLTSSIRR
Availability using Additional_repositories specification:
BOLTSSIRR no ?
? ? https://github.com/daviddaigithub/BOLTSSIRR
Additional repositories with no packages:
https://github.com/daviddaigithub/BOLTSSIRR
So the GitHub link does not seem to get recognized in the check. Might I have to change something here?
As you found, you can't use Remotes in a CRAN package. What you need to do is to make sure the .tar.gz file for the package you are depending on is available somewhere. Github doesn't do that automatically, because https://github.com/daviddaigithub/BOLTSSIRR isn't set up as a package repository.
The solution is to create your own small repository, and keep copies of non-CRAN packages there. The drat package (available here: https://github.com/eddelbuettel/drat) makes this easy as long as you have a Github account: follow the instructions here: https://github.com/drat-base/drat. In summary:
Fork https://github.com/drat-base/drat into your account, and clone it to your own computer.
Enable Github Pages with the docs/ folder in the main branch.
Install the drat package into R using remotes::install_github("eddelbuettel/drat"). (I assume this version will make it to CRAN eventually; if you use the current CRAN version instructions are slightly more complicated.)
Build the package you want to insert. You need the source version; you might want binaries too, if those are hard for your users to build.
Run options(dratBranch="docs"); drat::insertPackage(...) to insert those files into your repository.
Commit the changes, and push them to Github.
In the package that needs to use this non-CRAN package, add
Additional_repositories: https://yourname.github.io/drat
to the DESCRIPTION.
You will be responsible for updating your repository if BOLTSSIRR is updated. This is good because the updates might break yours: after all, it's still in development mode. It's also bad because your users won't automatically get bug fixes.
That's it, if I haven't missed anything!
EDIT: see edit in end.
I have a R package in github and I'm using devtools::install_github to install it, which also install dependency packages.
Recently this process will install httpuv as source package, but compiling it in Mac meet errors with automake (something like this). I installed automake, then there was error with clang: error: unsupported option '-fopenmp'.
The issue and the possible solutions 1 2 seemed to be quite complicated. I think the CRAN version of httpuv probably will work for me, and I don't want my users to go through so many errors and fixing compiler errors.
I'd like to just install all dependency packages from CRAN in binary. For some packages that do need the more up to date version, I have specified it in my package description with remote section.
I checked install_github, then install, then install.packages. It seemed that the default behavior for binary vs source package is
An alternative (and the current default) is "both" which means ‘use
binary if available and current, otherwise try source’. The action if
there are source packages which are preferred but may contain code
which needs to be compiled is controlled by
getOption("install.packages.compile.from.source").
My getOption("install.packages.compile.from.source") is interactive. This is actually a preferred behavior for me. However I never see the interactive prompt.
I tried to give a type = "binary" parameter in install_github, but it doesn't seem to work, maybe it's not passed to every dependency package install?
EDIT:
I found the situation is a little bit more complex:
my app specified to install shiny github version via remote in description. shiny specified to install httpuv github version in remote section too. So this is actually the intended behavior.
I'm not sure if there is a solution available, other than require CRAN version of shiny in my package.
EDIT 2: It's more complex than my previous findings.
I removed remote section in my package description, supposedly only CRAN version is needed. However install_github still install most dependencies from github.
I finally found out that I have these dependencies github version installed, so their description in my local disk have the github remote information, and install_github found this information and "upgrade" them again, even when some of them have no change.
So I need to uninstall them first, only use CRAN version.
The really problem here is that if a dependency package is already new, it should not be installed. It could be a bug of devtools.
install_github passes arguments to devtools::install, and there upgrade_dependencies= FALSE and maybe even dependencies = FALSE might be what you're after:
install_github("you/urPackage", upgrade_dependencies = FALSE)
I am making an R package with Rcpp. It works fine on my machine which has Rtools installed. But recently, I tried to install my package locally on a different machine (Windows) and got a compiling error. The reason was that on that machine there was no g++ compiler (for Windows, g++ is provided with Rtools). After installing Rtools, it worked just fine.
So the question is, if I upload it to CRAN, does it still requires users to install Rtools by hand? Or does the function install.package() detect and install Rtools for them?
Also, if you guys know some packages written with Rcpp, please let me know. I'd like to take a look how it works.
So the question is, if I upload it to CRAN, does it still requires users to install Rtools by hand?
No.
Or does the function install.package() detect and install Rtools for them?
No.
What happens is that CRAN builds pre-compiled binary files that can be installed by Windows and MacOS users without the need for compilers and related tools.
Also, if you guys know some packages written with Rcpp, please let me know. I'd like to take a look how it works.
rr <- devtools::revdep("Rcpp")
length(rr) ## 907
or see the Rcpp page on CRAN.
Users, e.g. people who download it via install.package(), are actually downloading a "compiled" version of the packaged called a binary that is maintained by CRAN. They will only ever need a copy of R.
On the other hand, Developers, e.g. people who are creating it, require development tools that are system specific. For those on the Windows platform, they must have a local install of Rtools on their machine. For developers on macOS, they must have their own copies of gfortran binaries and xcode developer line tools.
Lastly, there are many such Rcpp packages available to look to for inspiration...
See the Rcpp CRAN page
Pick a package and then look up the source at http://github.com/cran/packagename
View an annotated list by Dirk here.
My Fedora system (Fedora 20, all up to date) has just had R updated to version 3.1.0. Since then, I've had issues installing multiple packages. glmnet failed previously, and now I'm having trouble with treemap. More specifically, I get an error during treemap installation that httpuv has zero exit status.
I never had issues with the previous version of R. Any reason this version should have such problems??
There could be many causes to do with your OS, version, permissions, other installed packages/software, etc, etc. Without seeing the full error message it's hard to know.
One possibility specific to httpuv is root privileges. I've noticed a few threads on various forums when searching for installation errors with this package and Linux, many of them mentioning root v. non-root issues. In another case, libuv needed to be upgraded.
I encounter package installation problems daily and I have some more general work-arounds as well. Hopefully one of these will solve your problem.
Install the package from source
download.file(url="http://cran.r-project.org/src/contrib/httpuv_1.3.0.tar.gz", destfile = "httpuv.tar.gz")
install.packages("httpuv.tar.gz", type = "source", repos = NULL)
Install using devtools via GitHub if the package supports it
Install RTools and re-try your package installation
Install an older version of the package
If those above do not work, then I dig deeper by referring to advice given to me by a VP of IT in my company. These comments were made in reference to frequent package installation problems I encountered when switching from Windows to Solaris:
There are two types of install/make problems. Missing .h files
and/or missing .so/.a libs. The reason for these are multiple:
1.- the package that delivers these is not installed. This means that those files cannot be found anywhere in the /usr tree. Solution is
install right package, make sure the files are there
2.- the includes are not found by the install configurator. This means some environment variable or install option is not properly set (this
is our case for RODBC). Figuring out which variable to set is
challenging without looking at the package documentation [fortunately, documentation is not hard to find!]
3.- the libs are not in the LD_LIBRARY_PATH, easy to fix.
4.- There is a deeper compile/link error, meaning the package is not compatible with the rest of the sw, or has not been properly ported.
Some time ago, you used to be able to install the rcom package in R to use COM scripting (eg, access to external programs.) Unfortunately, it seems to be discontinued:
Package ‘rcom’ was removed from the CRAN repository.
Formerly available versions can be obtained from the archive.
This depends on statconnDCOM, which nowadays restricts use, contrary
to the CRAN policy for a package with a FOSS licence. See
http://rcom.univie.ac.at/ and http://www.statconn.com/.
Following the archive and statconn links and installing one of the older versions in R version 3 gives the error:
“Error: package ‘rcom’ was built before R 3.0.0: please re-install
it”.
I am not very familiar with R, but there seems no way around this message - after all, it occurs when installing, so re-installing doesn't seem to be the answer. It appears as though rcom is simply not available for recent (3.0+) versions of R. I have also scanned the package list, although searching for "COM" there returns over a hundred results and it is possible I missed the right one when clicking through them.
How can I use the rcom package, or use COM from within R some other way?
(Note: I am asking this question on behalf of a colleague. I have no experience with R myself at all. Both of us, when searching for answers, could not find anything. I am sure that others are also using COM in the latest version of R, though!)
I looked at the rcom source code a few months ago. It seems I can get it to build and install OK on R3.0.1. Below is the procedure if it helps.
Get a checkout of the latest source code of rcom. I have rcom_2.2-5.tar.gz locally. I can google something at the following address, but I have no idea of the provenance, so up to you to check it is legit. http://cran.open-source-solution.org/web/packages/rcom/index.html
in R do install.packages('rscproxy')
install Rtools as per the instructions on the R web site (http://cran.r-project.org/bin/windows/Rtools),
open a Windows command prompt i.e. run "CMD"
go to the folder containing the 'rcom' folder, and at the command prompt:
set R="c:\Program Files\R\R-3.0.1\bin\i386\R.exe"
%R% CMD check --no-manual rcom
check it passes without too many complaints. Your call as to the --no-manual option (if you have MiKTeX installed you may remove it)
%R% CMD INSTALL rcom
should result in
installing to c:/Rlib/rcom/libs/i386
** R
** inst
** preparing package for lazy loading
** help
*** installing help indices
** building package indices
** testing if installed package can be loaded
rcom requires a current version of statconnDCOM installed.
To install statconnDCOM type
installstatconnDCOM()
This will download and install the current version of statconnDCOM
You will need a working Internet connection
because installation needs to download a file.
* DONE (rcom)
in R:
library(rcom)
installstatconnDCOM()
I tried a comRegisterRegistry() ; comRegisterServer()
; x<-comGetObject("Excel.Application") but I get a NULL for x. I am not a user of rcom so while it all seems to compile fine; it may just not work anymore.
If you happen to need to access .NET code, a viable option (and yes I have a vested interest in mentioning it) may be the rClr package.
Hope this helps; I'd be interested to hear how you go.
This really should be a comment, but I don't have enough rep points yet to leave one.
I found that the above steps did not work for me, but the answer posted by Lisa Ann on this question, RExcel in R 3.0.x, did solve my problem installing rcom. Since you need rcom to run RExcel, the initial steps to install RExcel cover installing rcom on newer versions of R (such as 3.0.2).
Specifically, following the advice on statconn's wiki, http://homepage.univie.ac.at/erich.neuwirth/php/rcomwiki/doku.php?id=wiki:how_to_install
You also need to follow these instructions if you upgrade R, i.e. you install a new >release of R after you have installed RExcel.
Download the statconn DCOM server and execute the program you downloaded
Start R as administrator (on Windows 7 you need to right-click the R icon and click the >corresponding item)
In R, run the following commands (you must start R as administrator to do this)
install.packages(c("rscproxy","rcom"),repos="http://rcom.univie.ac.at/download",lib=.Library)
library(rcom)
comRegisterRegistry()
Now you have rcom installed, [instructions for installing RExcel follow...]
New versions of rcom and rscproxy (also for current versions of R) are available from a different repository. Just use http://rcom.univie.ac.at/download as the R repository to install from and you can download and install binary versions of statconn packages from there.
Hope this helps!