[R-package] fixed overly-strict compiler check in install.libs.R (fixes #2259)

[jameslamb]

As of #2164 , the R package picked up a fix in install.libs.R that handles difficulty finding OpenMP stuff when building on Mac OS Mojave.

Per discussion on #2259 , we found that that fix only works if environment variable CC is set to something that literally starts with gcc (and the same with CXX and g++). That could cause build failures if you pass a fully-qualified path like CXX=/usr/local/bin/g++-8. Thanks to @brandenkmurray for the help identifying that as the problem!

In this PR, I relax that check so that both g++-8, g++, and /some/path/to/g++-8 will all work. I did not take an opinion on the versions of g++ that are known to work with lightgbm (as suggested in this comment) as I think that would increase our maintenance burden by duplicating the number of places that is specified.

I tested that this works (lucky for us I'm running Mojave 10.14.5 on my laptop) and can confirm it's fixed. Code to test that it worked:

export CC=gcc-8 CXX=g++-8
Rscript build_r.R

export CXX=/usr/local/bin/g++-8 CC=/usr/local/bin/gcc-8
Rscript build_r.R

[StrikerRUS]

Oh, it's really useful fix! It seems that a lot of users are used to specify full paths in CC and CXX!

I guess that Python-package also has the same problem, so I'll propose the same fix for that in a separate PR.

The only thing I'm worrying about is possible collisions. I mean, isn't there any rare compiler name which contains gcc in its' name. For instance, clang and appleclang. According to the CMake docs and Wiki, there are no such names, but maybe I just should search better 😄.

[jameslamb]

I did find pgcc (http://www.dartmouth.edu/~rc/HPC/man/pgcc.html) on that Wikipedia page you linked! It's a compiler for C. It does seem like some people might be using it (https://stackoverflow.com/search?tab=newest&q=pgcc). I'll make the regular expression a bit more strict!

[jameslamb]

@StrikerRUS mind taking one more look? I sense that you were half-joking, but I do think guarding against name collisions is probably a good idea.

To check that the regular expressions I introduced here work, you can run these examples in an R session.

gcc <- 'gcc'
gcc8 <- 'gcc-8'
gcc8_fq <- '/usr/bin/whatever/gcc-8'
pgcc <- 'pgcc'
pgcc8 <- 'pgcc-8'
pgcc8_fq <- '/usr/bin/whatever/pgcc-8'

gpp <- 'g++'
gpp8 <- 'g++-8'
gpp8_fq <- '/usr/bin/whatever/g++-8'
whateverpp <- 'somethingg++'
whateverpp8 <- 'somethingg++-8'
whateverpp8_fq <- '/home/root/bin/somethingg++-8'

gcc_pattern <- '^gcc$|[/\\]+gcc|^gcc\\-[0-9]+$|[/\\]+gcc\\-[0-9]+$'

# should be true
grepl(gcc_pattern, gcc)
grepl(gcc_pattern, gcc8)
grepl(gcc_pattern, gcc8_fq)

# should be false
grepl(gcc_pattern, pgcc)
grepl(gcc_pattern, pgcc8)
grepl(gcc_pattern, pgcc8_fq)


gpp_pattern <- '^g\\+\\+$|[/\\]+g\\+\\+|^g\\+\\+\\-[0-9]+$|[/\\]+g\\+\\+\\-[0-9]+$'

# should be true
grepl(gpp_pattern, gpp)
grepl(gpp_pattern, gpp8)
grepl(gpp_pattern, gpp8_fq)

# should be false
grepl(gpp_pattern, whateverpp)
grepl(gpp_pattern, whateverpp8)
grepl(gpp_pattern, whateverpp8_fq)

[StrikerRUS]

LGTM! Thanks a lot for your research!
I don't think we need to consider some weird things in CC like gcconsole, which will trigger the regex.