Following #3971 and #3974
Dear CRAN,
Yes my latest submission a few hours ago should work on macOS now.
On Solaris it looks like it's compiling against version 1.2.8 of zlib headers (from previous output I saw of installation log) but using v1.2.11 at runtime (version in error message). I saw in the zlib manual that matching major version should be ok and that's checked at runtime by some zlib routines. But maybe that mismatch in version is not ok on Solaris for some reason. Unless, perhaps Prof Ripley has updated zlib in the meantime (the installation output would confirm as it prints zlib version there). Also, please can I see the full output not just the last 13 lines. It seems there are earlier errors which I can't see but should give me some clues. That is odd.
Since there are other errors fixed too (e.g. clang-ubsan) it would be good to allow the latest submission of 1.12.6 on CRAN to at least get binaries to macOS users, then I can come back to Solaris in v1.12.8. Assuming of course it does work on macOS now, and that Solaris is going to take longer.
Best, Matt
On Thu, Oct 17, 2019, 10:21 PM Kurt Hornik Kurt.Hornik@wu.ac.at wrote:
Matt Dowle writes:
On Thu, Oct 17, 2019, 1:18 AM Kurt Hornik Kurt.Hornik@wu.ac.at wrote:
Matt,
For the prev submission, Prof Ripley fouds the following:
This version does not install for me on macOS:
pkg-config did not detect zlib is installed. Please install it:
- deb: zlib1g-dev (Debian, Ubuntu, ...)
- rpm: zlib-devel (Fedora, EPEL, ...)
- brew: zlib (OSX)
ERROR: configuration failed for package ‘data.table’
macOS does not have pkg-config, nor does it supply .pc files for its
system software (SU has packaged some, but not zlib nor for current
macOS). And a pkg-config-only test is not portable for other platforms.
Can you pls fix accordingly?
Best
-k
Thanks, Kurt and Prof Ripley.
I don't assume that pkg-config is installed, but it appears then that it is
installed on that macOS box (as R-exts note 25 says) but, oddly, pkg-config
is claiming zlib is not installed when we know it is (since current CRAN
error with recent 1.12.4 on macOS is runtime error long after compilation,
which stands to reason since zlib is required by R itself). This all works
fine on Travis macOS so it is not macOS per se.
However, I can see a fix I can make for this situation; i.e., pkg-config
being installed but not returning an accurate result. I'll fix and resubmit.
I'm hoping this means the previous attempt works now on Solaris then? I can
see Rrof Ripley added 'checkbashisms' to R-exts note 24 as I suggested (so
he received that email), but I didn't hear back from him after that 2nd
Solaris attempt. (I didn't receive the reply you've quoted from him about
macOS.)
Prof Ripley had also told me about Solaris still not being fixed (see
below), but suggested we focus on the macOS installation problems.
Best
-k
Running ‘main.R’ [90s/48s]
Running the tests in ‘tests/main.R’ failed.
Last 13 lines of output:
x = fread(f1)
Empty data.table (0 rows and 2 cols): a,b
[Key= Types=log,log Classes=log,log]
y = DT
a b [Key= Types=int,int Classes=int,int]
1: 1 1
2: 1 1
---
199: 2 4
200: 2 4
Different number of rows
Error in fwrite(DT, file = f3 <- tempfile(), compress = "gzip") :
zlib v1.2.11 deflate() returned error -2 with z_stream.msg ''.
Please retry fwrite() with verbose=TRUE and include the full output with
your data.table bug report.
Calls: test.data.table -> sys.source -> eval -> eval -> fwrite
Following #3971 and #3974
Dear CRAN,
Yes my latest submission a few hours ago should work on macOS now.
On Solaris it looks like it's compiling against version 1.2.8 of zlib headers (from previous output I saw of installation log) but using v1.2.11 at runtime (version in error message). I saw in the zlib manual that matching major version should be ok and that's checked at runtime by some zlib routines. But maybe that mismatch in version is not ok on Solaris for some reason. Unless, perhaps Prof Ripley has updated zlib in the meantime (the installation output would confirm as it prints zlib version there). Also, please can I see the full output not just the last 13 lines. It seems there are earlier errors which I can't see but should give me some clues. That is odd.
Since there are other errors fixed too (e.g. clang-ubsan) it would be good to allow the latest submission of 1.12.6 on CRAN to at least get binaries to macOS users, then I can come back to Solaris in v1.12.8. Assuming of course it does work on macOS now, and that Solaris is going to take longer.
Best, Matt
On Thu, Oct 17, 2019, 10:21 PM Kurt Hornik Kurt.Hornik@wu.ac.at wrote:
Prof Ripley had also told me about Solaris still not being fixed (see
below), but suggested we focus on the macOS installation problems.
Best
-k
Running ‘main.R’ [90s/48s]
Running the tests in ‘tests/main.R’ failed.
Last 13 lines of output: