* Add continuous test for ruby 2.3, 2.4 and 2.5
* Change ruby 2.5 to 2.5.0
* No need to provide argument to rb_funcall when argc is 0
* Fix tests for ruby 2.5
* Use rescue instead of assert_raise to accept subclass of error
This avoids the need to use "yum update && yum upgrade" in the container
to be able to contact GitHub, which requires TLS 1.2[1].
I have verified that binaries built with this container still run in the
previous container; no errors like "/lib64/libc.so.6: version
`GLIBC_2.14' not found", which occur if using too new of a glibc when
compiling. CentOS 6.6 has glibc version 2.12 release 1.209.el6. CentOS
6.9 has glibc version 2.12 release 1.149.el6. Both would upgrade to
release 1.212.el6 via "yum update && yum upgrade".
1. https://githubengineering.com/crypto-deprecation-notice/
Now that we depend on C++11, we need at least GCC 4.8 instead of 4.7.
This change updates the Docker setup to continue using CentOS 6.6 but
with GCC 4.8.
I also added libm to the whitelist for dynamically linked libraries for
the ARM64 protoc binary.
1. Changed maven script to only do artifact uploading and removed build
script invocation from it. We didn't use maven to invoke the build
script before (we built protoc manually and editted pom.xml to only do
uploading for previous releases), and will not use it in the future (we
will use kokoro to build artifacts).
2. Cleaned up build-protoc.sh and README.md: removed the part about
using maven to build and listed supported platforms explicitly.
1. Changed maven script to only do artifact uploading and removed build
script invocation from it. We didn't use maven to invoke the build
script before (we built protoc manually and editted pom.xml to only do
uploading for previous releases), and will not use it in the future (we
will use kokoro to build artifacts).
2. Cleaned up build-protoc.sh and README.md: removed the part about
using maven to build and listed supported platforms explicitly.
Compilation of Python wrappers fails with Python 3.7 because
the Python folks changed their C API such that
PyUnicode_AsUTF8AndSize() now returns a const char* rather
than a char*. Add a patch to work around. Relates #4086.
