order of imports matters? about protobuf.js

[alexander-fenster]

Luckily, it turns out it's not actually about the import order, but about the fact that pbjs uses loadSync internally, and the "file not found" error gets swallowed in the sync mode, so the parsing continues and fails on another, seemingly unrelated, error.

The culprit in the example above was google/api/field_info.proto, which originally was the second import to load. It did not exist (for an unrelated reason), the error was thrown and the import processing loop was terminated on line 133 in the following snippet:

protobuf.js/src/root.js

Lines 120 to 141 in 4436cc7

	try {
	if (util.isString(source) && source.charAt(0) === "{")
	source = JSON.parse(source);
	if (!util.isString(source))
	self.setOptions(source.options).addJSON(source.nested);
	else {
	parse.filename = filename;
	var parsed = parse(source, self, options),
	resolved,
	i = 0;
	if (parsed.imports)
	for (; i < parsed.imports.length; ++i)
	if (resolved = getBundledFileName(parsed.imports[i]) || self.resolvePath(filename, parsed.imports[i]))
	fetch(resolved);
	if (parsed.weakImports)
	for (i = 0; i < parsed.weakImports.length; ++i)
	if (resolved = getBundledFileName(parsed.weakImports[i]) || self.resolvePath(filename, parsed.weakImports[i]))
	fetch(resolved, true);
	}
	} catch (err) {
	finish(err);
	}

The failed invocation of fetch already called finish once, that first invocation of finish cleared the callback, so when finish was called the second time from the catch block, it did not do anything, and the parsing continued (until the next failure).

Reordering the imports so that the missing one goes last actually helped because the missing proto only provided some options which, being missing, apparently did not break the parsing.

Uh oh, that was a fun problem to debug.