first crack at checking compiler versions from R

[eddelbuettel]

currently limited to g++

works reasonably well -- log from my Ubuntu 15.10 box below

edd@max:~$ g++ --version | head -1                                  # default
g++ (Ubuntu 5.2.1-22ubuntu2) 5.2.1 20151010
edd@max:~$ Rscript -e 'Rcpp:::compilerCheck("5.2.1")' 
[1] TRUE
edd@max:~$ Rscript -e 'Rcpp:::compilerCheck("4.6.0")' 
[1] TRUE
edd@max:~$ Rscript -e 'Rcpp:::compilerCheck("5.3.1")'               # what 16.04 has
[1] FALSE
edd@max:~$ CXX=g++-4.5 Rscript -e 'Rcpp:::compilerCheck("4.6.0")'   # old version
[1] FALSE
edd@max:~$ 

[JanMarvin]

int gccver(){
  return(  __GNUC__ * 10000 +  \
    __GNUC_MINOR__ * 100     +  \
    __GNUC_PATCHLEVEL__);
}

Isn't this what you want?

[eddelbuettel]

@JanMarvin: Thanks. In fact, @dcdillon made a similar point ("just compile g++ -std=c+11 hello.cpp"). But I wanted something a little simpler and quicker.

We do actually have some tests in Rcpp already:

• inst/include/Rcpp/platform/compiler.h tests for a variety of features needed around/before TR1
• inst/discovery/cxx0x.R is an old (and outdated) check for minimal standards version

so I don't think the last word has been said. Right now it is more about helping users right at the start of a build process, ie in configure. We had a number of unfortunate error reports when folks on older RHEL/CentOS attempted to build RcppArmadillo, and of course flamed out fabulously as g++-4.4 doesn't even buy you half a cup of joe anymore.

[eddelbuettel]

I'm going to merge this in, it is unlikely to cause harm as it isn't even exported yet.

[jjallaire]

It would be cool to have a variation of this which could be called from Makevars and which throws an error when the minimum compiler version isn't met. The error would have a nice clear message indicating that the issue is the compiler version.

[eddelbuettel]

Certainly. The bigger question really is whether we want

• full-blown compiler-based tests which actually test particular compiler features (a la platform/compiler.h in the sources)
• something lightweight to throw decent error messages either via Makevars (but then everybody has to modify Makevars again)
• or just build on this and call via Rscript from a shell script configure which people copy as is into their projects (if they currently do not use configure).

Making the error message nicer/more informative is easy. I was just happy when I realized that I could test g++ as well as the common environment variables from R and use R to compare versions strings correctly. What we have here is a start, and possibly 'good enough' to replace what I just did in RcppArmadillo.

[jjallaire]

I think that #2 (have folks add a line to Makevars) strikes the best
balance, especially since everyone already has a Makevars and this is
pretty much a boilderplate one-liner.

On Sat, May 14, 2016 at 12:27 PM, Dirk Eddelbuettel <
notifications@github.com> wrote:

Certainly. The bigger question really is whether we want

• full-blown compiler-based tests which actually test particular
  compiler features (a la platform/compiler.h in the sources)
• something lightweight to throw decent error messages either via
  Makevars (but then everybody has to modify Makevars again)
• or just build on this and call via Rscript from a shell script
  configure which people call?

Making the error message nicer/more informative is easy. I was just happy
when I realized that I could test g++ as well as the common environment
variables from R and use R to compare versions strings correctly. What
we have here is a start, and possibly 'good enough' to replace what I just
did in RcppArmadillo.

—
You are receiving this because you commented.
Reply to this email directly or view it on GitHub
#474 (comment)

[eddelbuettel]

How do you suggest to hook it in? It is not just an expansion as PKG_CXXFLAGS ?

Or expand to something which gets called and remains silent on success but emits loud and clear error messages on failure?

[jjallaire]

Yeah, I'd say silent on success and loud and clear error messages on
failure.

On Sat, May 14, 2016 at 12:33 PM, Dirk Eddelbuettel <
notifications@github.com> wrote:

How do you suggest to hook it in? It is not just an expansion as
PKG_CXXFLAGS ?

Or expand to something which gets called and remains silent on success but
emits loud and clear error messages on failure?

—
You are receiving this because you commented.
Reply to this email directly or view it on GitHub
#474 (comment)

[eddelbuettel]

I thought about that while going for a run and I am not so sure we can 'break' cleanly there. It is R calling make calling our (Rscript-based ?) snippet. I still think configure is cleaner. To be seen...