makes dependency on zlib optional

[jangorecki]

Follow-up of #3872
It is NOT meant to solve Solaris issue, but general issue when zlib is not available.

---

before PR

docker run -it rocker/r-ver:3.5.3 Rscript -e 'install.packages("data.table", repos="https://Rdatatable.gitlab.io/data.table"); data.table::fwrite(iris, "iris.gz")'

*** pkg-config is not installed.
*** Compilation will now be attempted and if it works you can ignore this message. In
*** particular, this should be the case on Mac where zlib is built in.
*** However, if compilation fails, try 'locate zlib.h zconf.h' and ensure the zlib
*** development library is installed :
***   deb: zlib1g-dev (Debian, Ubuntu, ...)
***   rpm: zlib-devel (Fedora, EPEL, ...)
***   There is a zlib in brew for OSX but the built in zlib should work.
*** Note that zlib is required to compile R itself so you may find the advice in the R-admin
*** guide helpful regarding zlib. On Debian/Ubuntu, zlib1g-dev is a dependency of r-base as
*** shown by 'apt-cache showsrc r-base | grep ^Build-Depends | grep zlib', and therefore
*** 'sudo apt-get build-dep r-base' should be sufficient too.
*** To silence this message, please ensure that :
***   1) 'pkg-config --exists zlib' succeeds (i.e. $? -eq 0)
***   2) 'pkg-config --libs zlib' contains -lz
*** Compilation will now be attempted ...
...
gcc -I"/usr/local/lib/R/include" -DNDEBUG   -I/usr/local/include  -fopenmp -fpic  -g -O2 -fstack-protector-strong -Wformat -Werror=f
ormat-security -Wdate-time -D_FORTIFY_SOURCE=2 -g  -c fwrite.c -o fwrite.o
fwrite.c:10:49: fatal error: zlib.h: No such file or directory
 #include <zlib.h>      // for compression to .gz

---

after PR

docker run -it rocker/r-ver:3.5.3 Rscript -e 'install.packages("remotes", repos="https://cloud.r-project.org"); remotes::install_github("Rdatatable/data.table@nozlib"); data.table::fwrite(iris, "iris.gz")'

*** pkg-config is not installed.
*** Compilation will now be attempted and if it works you can ignore this message. In
*** particular, this should be the case on Mac where zlib is built in or pkg-config
*** is not installed. However, if compilation fails, try 'locate zlib.h zconf.h' and
*** ensure the zlib development library is installed :
***   deb: zlib1g-dev (Debian, Ubuntu, ...)
***   rpm: zlib-devel (Fedora, EPEL, ...)
***   There is a zlib in brew for OSX but the built in zlib should work.
*** Note that zlib is required to compile R itself so you may find the advice in the R-admin
*** guide helpful regarding zlib. On Debian/Ubuntu, zlib1g-dev is a dependency of r-base as
*** shown by 'apt-cache showsrc r-base | grep ^Build-Depends | grep zlib', and therefore
*** 'sudo apt-get build-dep r-base' should be sufficient too.
*** To silence this message, please ensure that :
***   1) 'pkg-config --exists zlib' succeeds (i.e. $? -eq 0)
***   2) 'pkg-config --libs zlib' contains -lz
*** Compilation will now be attempted ...
R CMD SHLIB supports OpenMP without any extra hint
*** Compilation without compression support in fwrite
...
Error in data.table::fwrite(iris, "iris.gz") : 
  Compression in fwrite uses zlib library which was not installed at the time when data.table was compiled. To enable compression support in fwrite one must install zlib and re-install data.table.

[MichaelChirico]

doesn't this turn off zlib support on mac by default

[jangorecki]

If pkg-config is not available on Mac then yes, it turns off. Needs testing anyway.
There is a related comment above:

if pkg-config is not available then zlib will be disabled for higher chance of compilation success

[jangorecki]

Note that this verbose message branch was hit before when pkg-config was not installed, just the message was not mentioning that.

[jangorecki]

We could eventually try to check if zlib is available by attempting to compile dummy code, like we do for openmp. Docker image in your report didn't have neither pkg-config and zlib. So then we could still compile with zlib support. For now I don't think it is worth. Messages are clear and points users in the proper direction. Installing pkg-config and/or zlib.

[codecov]

Codecov Report

Merging #4844 (08a2584) into master (354cf84) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #4844      +/-   ##
==========================================
- Coverage   99.47%   99.47%   -0.01%     
==========================================
  Files          73       73              
  Lines       14560    14559       -1     
==========================================
- Hits        14483    14482       -1     
  Misses         77       77              

Impacted Files	Coverage Δ
src/utils.c	98.36% <ø> (ø)
src/fwrite.c	97.69% <100.00%> (-0.01%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 354cf84...08a2584. Read the comment docs.

[jangorecki]

I put it on label High so we can have it merged at the beginning on a development cycle to be well tested before it will land on CRAN.

[mattdowle]

Changes I made in 08a2584:

• zlib runtime is still a requirement for those installing the binary package from CRAN. So reinstated the SystemRequirements zlib line in DESCRIPTION. Not a problem because the zlib runtime is already required by R. Further, test.data.table() will fail, and therefore so will R CMD check, if data.table has not been compiled with zlib headers because tests of fwrite compression will fail. So leaving that line as-is seems more appropriate than removing that line.
• #define z_stream struct line removed. I'm not sure that would result in valid syntax. I checked all uses of z_stream are inside #ifndef NOZLIB blocks which is stricter.
• distinction between library and headers made in the messages
• zlib headers may be installed but not found for some reason. So strictly speaking, using the words 'not found'.
• the error message now tells the user that reinstalling will output further guidance
• if (nth>1) verbose=false also removed (unrelated to this PR) as I neglected to do that in Remove tracing of zlib/fwrite for Solaris #4860. The comment on that line was clear that it was only there for the Solaris tracing.