Test against more relevant versions of node on Travis

[JaKXz]

allow for node@unstable failures

• slightly quicker npm install

[JaKXz]

Which of these compilers would be preferred?

[bcoe]

@JaKXz I'm cool with dropping iojs and upgrading to 6, but since nyc is aiming to be fairly backwards compatible could you leave node@0.10.

[JaKXz]

@bcoe done. I'll leave the Travis warning about GCC compilers to you / the other maintainers if there's some preferred one necessary - or I can just add whichever you guys want. I don't know the differences well enough to make a substantial decision.

[novemberborn]

Does this do anything? npm won't render a progress bar in Travis anyway.

[JaKXz]

@novemberborn I think it does speed up install a bit, it definitely does locally... I don't think there's harm in having it on CI since like you said Travis doesn't render one anyway, but if there's even a slight benefit then it's worth it imho.

[novemberborn]

I don't think there are any benefits for Travis (and I doubt whether there's significant ones locally). Best not to have invocations like this in the code, it'll only lead to more questions in the future.

[novemberborn]

Should leave 0.12 as well, it's supported to the end of the year (and even afterwards we should refrain from unintentionally breaking compatibility). I'd prefer testing 5 while it's still being maintained.

[bcoe]

@JaKXz thanks for entertaining our 🚲 🏠, will land this today :)

[JaKXz]

@bcoe TBPH I don't mind at all because this is really a discussion for you guys to have as maintainers (about which versions you're supporting etc) - I just noticed that 1) the builds were really slow and 2) that the versions of node were [well] out of date so I thought I'd kickstart things :)

[JaKXz]

@novemberborn this and the continuing discussion at npm/npm#11283 are why I suggest the before_install command - I don't mind being vetoed here but I don't think it's harmful.

[bcoe]

@JaKXz easy enough to test; let's kick off a build with and without the setting, and look at the Travis benchmarks; there's a second count beside each step in the build.