fix: updated isNode check (#1221)

[diegoblattner]

• Fixes util.isNode / Error: not supported #1221 when used within SSR.
  process is a global variable since node 0.1.7 and versions since 0.2.0.

[dcodeIO]

Haven't checked and might be a false alarm, but referencing process directly as proposed here might cause the bundler to pull in a process shim, increasing distribution size, and potentially breaking browser builds due to assuming they'd run under node. Thought I mention :)

[alexander-fenster]

Thanks @dcodeIO, let me check this. This is basically why we decided to merge stuff on Fridays and let it bake for a while before we make an actual release next week :) I'll update the issue and/or send a fix if needed.

[alexander-fenster]

So, several things here:

• it indeed includes a shim for process now (I checked with webpack). It pulls in the whole ./node_modules/process/browser.js which adds 1.6K of code when minimized. That's sad.
• it may or may not break some builds indeed (if someone just checks for typeof process !== "undefined" to check if it's Node or not, this code may stop working within the bundle), even though this check is actually wrong;
• adding a shim for process can be disabled in the bundler config (e.g. webpack.config.js or similar).
• there is an NPM module called browser-or-node (28.5k weekly downloads) that actually implements isNode in the same way as introduced in this PR.

My personal opinion here is that it's better to fix the code for users who actually reported they have problems (in #1221), and then fix it once more if the fix actually breaks someone else's code (it definitely will, obligatory xkcd). Folks who really care about the extra 1.6K of minimized code will be able to configure their bundler to remove this shim.

[alexander-fenster]

...but the bad thing is that it actually breaks npm run make (ReferenceError: process is not defined), so yes we'll either need to rollback or figure out a different way to fix it.

[dcodeIO]

The util.global alias was intentionally introduced to avoid the shim bundling issues we are having now, so perhaps #1221 (comment) is a better fix than this one. Like, util.global.process being missing appears to be more of a symptom of an underlying issue with what we alias into util.global. If the fix proposed within the linked comment doesn't introduce any immediate issues, I recommend to revert this PR and fix util.global as proposed instead.

[dcodeIO]

Perhaps with an additional safeguard there:

function checkNodeGlobal(global) {
  if (global && global.process) return global; // or whatever reasonably identifies node
  return null;
}

// global object reference
util.global = typeof global !== "undefined" && checkNodeGlobal(global)
           || typeof window !== "undefined" && window
           || typeof self   !== "undefined" && self
           || this; // eslint-disable-line no-invalid-this