* external/com_google_protobuf/src/google/protobuf/stubs/strutil.cc:1122:9: runtime error: negation of -9223372036854775808 cannot be represented in type 'google::protobuf::int64' (aka 'long'); cast to an unsigned type to negate this value to itself

* Bad external/com_google_protobuf/src/google/protobuf/text_format.cc:1320:14: runtime error: null pointer passed as argument 1, which is declared to never be null
  /usr/include/string.h:62:62: note: nonnull attribute specified here
Signed-off-by: Harvey Tuch <htuch@google.com>

Merge 3.7.x into master

Also update Dockerfile for JavaScript. The old one is using an npm that is too outdated

This actually has always been using the first branch because
`__cpluscplus` is not a real definition (contrast with the real spelling
of `__cplusplus`).

* add wrapper type helpers

* add check for _as_value suffix

* add implicit time conversion

* add duration

* add init test

* more tests

* add type check and alternative c type check

* add rational and bigdecimal

* use rb_obj_is_kind_of

* use native time check

* chain implicit conversions

* remove unused variable

I've deprecated the separate web page for the project and moved everything to `README.md`, so for now it's better to point to the GitHub project page.

* Update version number to 3.7.1

* Update version number in port_def.inc

OSX has deprecated OpenSSL, so we need to install it via Homebrew and
build against that version when we build Ruby.

We ran into problems with precompiled protoc binaries not working on
some systems because of the dependence on a dynamically linked
libatomic. However, I found that linking against libatomic is not
actually necessary in our protoc build environment. Inside configure.ac
there's a test which figures out if we need to pass -latomic, and it
does this by attempting to build a test binary. That binary was failing
to build, but it turns out this was because we need to pass -std=c++11,
not -latomic.

This fixes #5875.

* Allow bytes field to be longer than 16000 bytes

* Remove empty line

We need to mark messages as final soon, but before we do that we need to
provide a temporary opt-out mechanism to accommodate existing code that
inherits from generated messages. For 3.7.1 let's stop marking messages
final but in 3.8 we can reintroduce "final" with an opt-out option.

Down integrate to Github

The removed variable here is actually used breaking downstream bazel
builds.

Signed-off-by: Mickaël Schoentgen <contact@tiger-222.fr>

The bazel-bin/ and bazel-genfiles/ directories are going to be merged
soon. That change was causing some test failures on Windows, and I
believe it is because the logic for locating test data files was getting
confused and mistakenly looking for them under bazel-bin/. This commit
updates that logic to look for a more specific file (descriptor.cc)
which does not appear under any Bazel-related directories.

It sounds like the more official solution is to use the Bazel runfiles
library (i.e. //tools/cpp/runfiles). However, I decided not to do that
because we currently still support multiple build systems, and if we
used a separate solution for Bazel then I suspect we would need even
more #ifdefs in the code to handle the different systems.