Null "browser" values appear to break downstream webpack builds

[jeffbaumes]

In the candela library, which depends on vega and uses webpack, we started seeing the following error earlier today:

Undefined identifier
/Users/jeff/code/candela/node_modules/enhanced-resolve/lib/Resolver.js:184
	var idxQuery = identifier.indexOf("?");
	                          ^

TypeError: Cannot read property 'indexOf' of null
    at Resolver.parse (/Users/jeff/code/candela/node_modules/enhanced-resolve/lib/Resolver.js:184:28)

It is being triggered by some logic in enhanced-resolve when attempting to load canvas from vega-canvas. The value of data is null (not undefined or false), so it passes the checks and gets sent on to lower stacks until it is eventually referenced and gets the indexOf error (doRequest() is called with the request field set to null):

 31 		var data = typeof data1 !== "undefined" ? data1 : data2;
 32 		if(data === innerRequest) return callback();
 33 		if(data === undefined) return callback();
 34 		if(data === false) {
 35 			var ignoreObj = Object.assign({}, request, {
 36 				path: false
 37 			});
 38 			return callback(null, ignoreObj);
 39 		}
 40 		var obj = Object.assign({}, request, {
 41 			path: request.descriptionFileRoot,
 42 			request: data
 43 		});
 44 		resolver.doResolve(target, obj, "aliased from description file " + request.descriptionFilePath + " with mapping '" + innerRequest + "' to '" + data + "'", createInnerCallback(function(err, result) {

I don't know anything about the browser field in package.json, but this spec seems to suggest using false, and does not mention the use of null. Should nulls be changed to false or undefined in package.json instead to make tools like webpack happy?

That's the best analysis I can do with limited knowledge on the subject.

[domoritz]

This is also causing build issues in ipyvega and jupyter renderers.

[domoritz]

However, I can compile successfully with webpack 1. #jsproblems