Rimraf npm

[MylesBorins]

build: remove npm on make install

Different versions of npm potentially have different trees. We should
be cleaning the version of npm on every install.

This implements the removal naively by calling
rm -rf $PREFIX/lib/node_modules/npm in the make install command.

This does not implement similar functions for the windows installer,
nor does it implement the removal in the double click installers.

This can be implemented as part of this PR or as part of another.

[jbergstroem]

Doing rm -rf $var/foo will get unixbeards turning in their grave. Suggesting you wrap this with a few checks if you insist on doing so.

[MylesBorins]

@jbergstroem I may or may not be implementing things in a naive way to get the ball rolling 😄

[MylesBorins]

Also worth mentioning that if we can hit all the edge cases (all installers) we should be able to close #3606

[mscdex]

I turned over in my grave and I'm not dead yet.

[Fishrock123]

@thealphanerd This isn't the correct fix though. I've never had problems via the Makefile. The installers need to be doing this.

[MylesBorins]

@Fishrock123 I've had this problem turn up with the makefile as well when moving between LTS releases.

I'm part way through a fix for the osx installer

[jbergstroem]

@thealphanerd: perhaps look at implementing this in tools/install.py instead? Would give a few way of making sure that path is correct (and potentially verify the structure).

The question whether deleting the folder or not still remains. Are the edge cases causing installer issues known?

[wacky6]

OSX 10.11, upgrading from v5.10 to v6.0 using official OSX Installer.

This problem still exists.

rm -rf /usr/local/lib/node_modules/npm then reinstall v6.0 solves the problem.

[bnoordhuis]

This doesn't do what you think it does, the path gets evaluated to 'REFIX'/lib/node_modules/npm. I don't want to sound harsh but did you actually test it?

[MylesBorins]

tbqh I was whipping this up while doing a handful of other things. I assumed the rimraf implementation would not be accepted and wanted some suggestions on next steps. I've updated with ${PREFIX) so that it actually works, and will explore doing this in the python script

[bnoordhuis]

I'd be more inclined to LGTM this if it was a pre- or post-install pass in tools/test.py that checks whether the install directory contains unexpected files. Or perhaps just a warning if the directory already exists.

[MylesBorins]

I'm going to close this for now, I'll revisit later