Fix util.global

[dcodeIO]

An alternative to #1363, attempting to fix the problem at the util.global level as proposed in #1221 (comment).

Reverts #1363 and #1440, with the latter not being needed anymore. See also: #1363 (comment)

[alexander-fenster]

Thank you @dcodeIO, you definitely have more experience with bundlers than I do :)

[dcodeIO]

Sorry for second-guessing btw :). If I have any bundler experience, then it is in preventing them from functioning as intended. As lost as everybody else otherwise.

[evgenius]

Object.prototype.toString.call(global) === "[object global]" seems to be too strict.
Now protobufjs fails to initialize in the Jest environment since Jest modifies global. See https://github.com/facebook/jest/blob/d7f3427/packages/jest-environment-node/src/index.ts#L30

[dcodeIO]

Hmm, more forgiving checks there could be

Object.prototype.toString.call(global.process) === "[object process]"

if process remains unmodified, and/or

global.setTimeout === setTimeout

to just validate that it's at least somewhat an actual global object.

[dcodeIO]

A more general question would be what Jest is actually trying to test there. Does a Jest test suite automatically assume that it's testing in a node environment, or may it also test client code while filling in browser APIs? If the latter is possible, protobuf.js cannot decide on its own but will need some sort of flag to indicate, i.e. setting util.isNode manually.

[evgenius]

How about changing the order?

util.isNode = typeof global !== "undefined"
           && global.process
           && global.process.versions
           && global.process.versions.node;
util.global = util.isNode && global 
           || typeof window !== "undefined" && window
           || typeof self   !== "undefined" && self
           || this;

If something tries to pretend it's Node by having versions.node, I think it should be just allowed to do so. What do you think?

[dcodeIO]

That looks good as well, just needs another check && global before accessing global.process in case it's null.

[evgenius]

Ah, right, forgot about `null` :)

[alexander-fenster]

Folks also suggested #1444 as a possible fix to the strict [object global]. I must admit I lost track of all the possible environments and frameworks this code should work in :) But I would really appreciate if all the fixes for "does not work for me" cases try to update the test suite if possible, otherwise we will keep breaking stuff (with good intentions of course).