The `brew test` tests were failing as they were unable to include
`test/unit/assertions`. This is because it's a gem and we were setting
the `GEM_HOME` so system gems were being ignored.
While I was there and examining the `$LOAD_PATH`: reduce the number of
things we add there by using `Bundler.with_clean_env` and only adding
`HOMEBREW_LIBRARY_PATH` in the `$LOAD_PATH` if it isn't already there
(which it always is [and has to be] for integration tests).
This also seems to have the side effect of speeding up integration
tests from 1m26s to 1m8s on my machine.
- Avoid using a temporary variable where not necessary
- Use fewer, better stubs in the tests to avoid warnings and better
test the implemented functionality.
instead of writing a .npmrc file, which simplifies the code.
npm_cache_config is still preserved for backwarts compatiblility and
usage int he kibana@n formulas in core.
This tells npm pack to don't run prepublish scripts at all.
I think this is the best default because:
* most modules don't have a prepublish script at all and aren't affected
by this change
* most prepublish scripts are calling devDeps, which would fail in our
case, because (dev)Deps aren't installed at npm pack time until #2820
gets resolved
* we favor npm registry tarball for formula downloads, which are already
prepublished, so we would in the best case needlessly run prepublish
a second time and in the worst case it would fail (because a clean
step is required before running prepublish a second time in a row)
* This change does the right thing for >99% of all the packages and
would only affect packages with prepublish scripts downloaded from a
non-npm registry tarball (like github tarballs) and with a prepublish
script wich does no't require any devDep (unlike for cross platform)
By telling node-pre-gyp and prebuild to don't pull prebuild binaries and
instead build them from source. This still may not work for some custom
third party scripts for pulling prebuild binaries.
