doc: update g++ requirement from 4.8 to 4.8.5

[bnoordhuis]

Lest people think that g++ 4.8.0, released in March 2013, is acceptable.
4.8.5 was released in June of this year and contains numerous bug fixes.

R=@misterdjules?

[rvagg]

is this just because we can or are you aware of bugs in .0 -> .4 that are serious enough to warrant this as a strong recommendation?

[bnoordhuis]

The context is #3391 (comment). It made me realize that using an old 4.8 release is potentially troublesome.

[cjihrig]

LGTM

[misterdjules]

According to GCC's ABI Policy and Guidelines, that would make the minimum C++ runtime requirements jump from GLIBCXX_3.4.18, CXXABI_1.3.7 to GLIBCXX_3.4.19, CXXABI_1.3.7.

So I think we would first need to:

1. Demonstrate that no currently supported platforms will break
2. There's an actual specific problem with using any other g++ 4.8.x version

[rvagg]

@bnoordhuis I see the link to the gcc bugzilla which is interesting but doesn't highlight any specific bugs that would impact on builds.

Also, RHEL's devtoolset-2, the CERN variant of which we are using to compile the Linux binaries on CentOS5 only comes with 4.8.2. gcc version 4.8.2 20140120 (Red Hat 4.8.2-15) (GCC) to be exact. So -1 on bringing the requirements up even further than what we use ourselves.

[Fishrock123]

I see no problem in doing this for the master branch?

[misterdjules]

@Fishrock123 I believe we'd have the same concerns for upgrading GCC requirements in the master branch than in the current LTS branch. If as a result the node binary doesn't run on some platforms we currently support, then the impact of that change should be at least evaluated and documented properly, and advance warning should be given to users of these platforms.

[Fishrock123]

If as a result the node binary doesn't run on some platforms we currently support, then the impact of that change should be at least evaluated and documented properly, and advance warning should be given to users of these platforms.

I don't disagree, but consider if v8 uses something that bugs out in an older compiler, what do we do?

[misterdjules]

@Fishrock123

I don't disagree, but consider if v8 uses something that bugs out in an older compiler, what do we do?

That's an excellent question, but it is outside of the scope of this PR, unless we know of such a problem today. I suggested the following in #3391 (comment):

It seems that it would help to have a detailed matrix of the runtimes/OSes/platforms supported by the binaries available on nodejs.org. We could use that to communicate potential breakage when we want to bump the compiler/runtime requirements, or at least to understand exactly how bumping the minimum required compilers would impact the list of supported platforms.

[benjamingr]

Status?