build: Improve the configure experience

[jbergstroem]

Not sure about the PR topic here, but a few - imo - important changes in here, mainly concerning shared builds:

• All handling of headers/libraries/etc for a shared library is only processed if a shared build is enabled (--shared-foo)
• I incorrectly handled return from b() in a previous commit. We now compare strings instead.
• Move shared and i18n flags into their own option groups with a short description of each
• i18n --with-intl is now validated the same way other choice options are (optparse)
• better integrate with pkg-config if shared builds are enabled and let users override pkg-config path via the environment (useful in cross-compilation environments, found in a floating patch)

Open to further suggestions to improve the language of configure which is decent at best. Also, let me know if I should squash (and how).

/R=@iojs/build, ... ?

Edit: Adding R=@bnoordhuis since he regularly touches build/configure.

[jbergstroem]

In the last @iojs/build meeting we talked about potentially checking bundled versions with pkg-config. I think it's a good thing since it helps us in cases where we assume some versions being available (we will soon depend on openssl 1.0.2-specific stuff and we already assume libuv is at least 1.0).

The biggest downside of this would be the complexity added to configure for parsing version out of bundled deps. I don't really see adding static versions somewhere as an alternative.

How does others feel about this?

[monsanto]

Is there some reason for writing if b(expr) == 'true': instead of just if expr:?

[jbergstroem]

@monsanto since b() always returns a string (its first usage was probably to return a gyp-friendly string to put into our generated config.gypi), python will treat that statement as boolean true:

>>> foo = 'false'
>>> if foo:
...   print 'true after all'
... 
true after all
>>> if foo == 'true':
...   print 'you\'ll never see me'
...

[monsanto]

@jbergstroem I don't understand what your point is. if b(expr) == 'true': and if expr: are exactly the same by the definition of b(), the only reason to prefer the former is some weird in-house style (which is why I'm asking).

[jbergstroem]

@monsanto sorry, I misunderstood the comment. Your suggestion is correct -- just pushed a new version.

Edit: simplified my reply (no need to elaborate on reasoning).

[rvagg]

lgtm, might be good to get sign-off from @bnoordhuis too

[jbergstroem]

So, current state:

• drop parts of 867de71 since it already landed
• @bnoordhuis wants --shared-whatever-includes to override --shared-whatever (--shared-whatever-libraries is already set by default) as convenience. I'm not very keen on this but open to discuss.
• I'll do some more work on cases that should fail, such as mixing shared and static flags.
• I'd like to better fence around what we assume as default from pkg-config -- even if pkg-config will likely provide more accurate information, it would just overwrite the user input based on this patchset (and already does for openssl in master/v1.x)

[jbergstroem]

Ima close this and split it up into different PR's instead. Easier to review and discuss.