JBIDE-20351: Improving Node detection for Ubuntu

[ibuziuk]

No description provided.

[alexeykazakov]

What will happen if there are both "node" (not a node.js package) and "nodejs" (node.js package) available? Which one will be used? Incorrect "node"? If so, shouldn't we look for nodejs on Linux first and if we fail then look for "node"? Is there any good way to distinguish a "fake" node and a real one?
And BTW we need to add test for nodejs detection and for "node" too if we don't have it already.

[ibuziuk]

What do you mean by "fake" ? Both "nodejs" and "node" possible names for Node.js executables

[maxandersen]

yes, but on ubuntu it seems you can have both node and nodejs available at the same time and node won't be the right one.

[alexeykazakov]

Exactly. "node" in Ubuntu is usually a command from Amateur Packet Radio Node Program package. Not from Node.JS. It's why they use nodejs. So in Ubuntu, if you have both "node" and "nodejs" then it's more likely that "nodejs" is the right one and "node" is the wrong one.

[ibuziuk]

Ok got it. Will update PR tomorrow. Need to figure out how to distinguish two of those executables

[alexeykazakov]

Don't forget to add tests for "nodejs"/"node" detection ;)

[ibuziuk]

got an idea of how to distinguish Amateur Packet Radio Node from Node.js executables. Basically, those two have a significant installed size difference - Amateur Packet Radio Node ~ 34Kb ( http://packages.ubuntu.com/trusty/node) and Node.js ~ 7 - 11 Mb depending on version. So, depending on the file's metadata we can find out whether "node" is trustworthy. WDYT ?
In terms of test, should I mock node locations?

[alexeykazakov]

What I see in your code. If there are both "nodejs" and "node" on Linux then "nodejs" is used. If there is only "node" then you use it only if its size more than 5MB.
But I guess this won't work if user created a symlink named "node" to the "nodejs". The size of this link will be small. You can resolve symlinks using a canonical path. And I would decries this limit of "node" files to something insane for a real node.js binary. 300KB or something like that. So if they shrink some feature version of node.js we will still be able to detect it. IMO it's better to detect a wrong "node" (it's a very well known problem for node.js users in Ubuntu) than to not recognize a "node" tuned by some advanced user.
And sorry for being a pain in the butt but I'm still waiting for tests ;)

[ibuziuk]

tests are not at all a pain in the butt ;-) I just need a piece of advice how to write good tests for this particular case. I mean how this public detectNode() method can be properly tested ?
Today I figure out that we might probably over complicate this problem with Amateur Packet Radio Node program. Here are my thoughts:

1. if we detect nodejs first on linux there will be no problem because...

2. Let's assume that user know that native node must be pre-installed. Than there is no problem - he will either install nodejs ("nodejs" binary) our install nodejs-legacy ("node" binary). If he had this Amateur Packet Radio package and still want to use Node.js as "node" command he will have to perform apt-get remove node and remove Radio package node. As you said, this is a well-known issue. details here - http://askubuntu.com/questions/235655/node-js-conflicts-sbin-node-vs-usr-bin-node

3. If the user doesn't know that native node must be preinstalled. If there is no Amateur Packet Radio package on his ubuntu, we are fine - pop-up will be shown with info that node.js was not detected, so now he aware of the fact that he must pre-install Node.js . if he has this package and we have detected it, once "bower update" shortcut will be executed, he will get pop-up that launch was not successful and preferences with node / bower might be the case. Once he will figure out that he must install nodejs we will have no problem.

[ibuziuk]

and one more point here: in order to install native bower user will need to install node.js and npm firstly. So, if he has only Amateur Packet Radio node the error pop-up for "bower update" will be shown anyways, cause bower will not be detected. Once node.js, npm and bower installed the detection will work just fine.

[alexeykazakov]

OK that makes sense. So just using nodejs if it's available will solve the problem. If it's not available then we should look for "node". That's it.
And I will think about tests.