Cannot read property 'readFileSync' of undefined

[murrayju]

Everything works fine with CoffeeScript v1.10, but if I drop in v1.11 or newer I get this error in both the browser (using require-cs plugin) and in my r.js (requirejs) build.

Cannot read property 'readFileSync' of undefined

If I debug this in the browser (sorry for the minified code):

f._compileFile = function(c, d, f) {
    var g, k;
    null == d && (d = !1);
    null == f && (f = !1);
    k = v.readFileSync(c, "utf8");

with a breakpoint on the last line above, v is undefined.

Let me know what else I can provide to help debug this.

[GeoffreyBooth]

Please provide your original coffee code. Also, require('fs') shouldn't work in a browser.

[murrayju]

Ok, I've made a plnkr that reproduces this:
https://embed.plnkr.co/WnoV7Q7MiB7CZ505FdAp/

Run that with the chrome debugger open, and you'll see the Uncaught TypeError that happens in coffee-script.js. I'm using the "browser-compiler" available at https://raw.githubusercontent.com/jashkenas/coffeescript/master/docs/v1/browser-compiler/coffee-script.js

I did a bit of digging, and it seems that this is only a problem when there is an Error stack trace that contains a reference to a .coffee file. Specifically, you have to evaluate e.stack to cause the uncaught TypeError to occur.

[GeoffreyBooth]

Error.stack is nonstandard: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Error/Stack. I don’t know why you would be using it in production code.

[murrayju]

It is not my code, it is actually in kriskowal/q. They are doing some feature detection to see if the browser supports stacks, and it is this detection code that invokes e.stack which in turn invokes your code, which throws an exception.

Anyways, I think this is a moot point. There is a code path in your "browser-compiler" build that will try to invoke require('fs'), which is obviously a bug.

In the meantime, I can probably fork q as a workaround.

[GeoffreyBooth]

parser.js, which is generated by jison, has the following at its end:

if (typeof require !== 'undefined' && typeof exports !== 'undefined') {
exports.parser = parser;
exports.Parser = parser.Parser;
exports.parse = function () { return parser.parse.apply(parser, arguments); };
exports.main = function commonjsMain(args) {
    if (!args[1]) {
        console.log('Usage: '+args[0]+' FILE');
        process.exit(1);
    }
    var source = require('fs').readFileSync(require('path').normalize(args[1]), "utf8");
    return exports.parser.parse(source);
};

Which is clearly supposed to only run in a CommonJS environment, but presumably your require.js setup is confusing it. I suggest you shadow require within the function that includes CoffeeScript, so that require is undefined when this code checks it.

This code isn’t part of CoffeeScript anyway, it’s jison, so it’s not in our domain to change.

[murrayju]

This is also very similar to #3890. I may try to implement something similar to atom/coffee-cash#1 as a workaround.

Seems reasonable to ask that CoffeeScript's Error.prepareStackTrace should not throw exceptions.

[GeoffreyBooth]

Please open a ticket against jison. That’s the module generating the problematic code.

[murrayju]

@GeoffreyBooth I'm not trying to be difficult here, but I find your stance frustrating. The fact that you are using jison is an implementation detail. From the user's perspective, CoffeeScript has a bug, and it is up to you guys to fix it.

Let's say that this is fixed in jison tomorrow. CoffeeScript would still have this bug at least until the new version of jison is pulled in and CoffeeScript is recompiled. The burden is still on you to track this and make sure that it gets fixed.

While I could try to open a ticket against jison, this would be difficult for me because I know nothing about jison, or how CoffeeScript is using it. It would make much more sense for you to create this ticket, since you have all of this information.

There is also a simple temporary workaround measure that could be implemented in the CoffeeScript source. Just add a try-catch around the function here: coffee-script.coffee#L352. I could make a pull request that does this if you'd like, but I don't want to waste my time if you won't merge it.

[GeoffreyBooth]

I would be open to accepting a PR, but I’m not familiar with that part of the codebase. It would need to be approved by @lydell or @jashkenas.

[murrayju]

@GeoffreyBooth @lydell @jashkenas see the pull request #4399. This totally solves this issue for me, but is more of a workaround. You may want to consider fixing the root of the problem (with jison?) at some point, but I think this would be fine for now.

[davibe]

The "fix" removed an important feature making all stack traces useless in my backend implementations. #4418