Planck and writing tests

January 29, 2016

Lately I’ve been playing around with Planck, a commandline bootstrapped ClojureScript REPL for OS X based on JavaScriptCore.

There are two reasons for this, one, I’m trying to write as much Clojure(Script) as possible and one way to do that is to try to automate as much as possible using Clojure(Script), and two, it’s fun to contribute to some opensource projects.

Planck consists, as far as I can tell, of a wee bit of Clojurescript and quite a bit of Objective C which lets you augment JavasScriptCore with functionality which isn’t readily available.

Since Planck has fantastic startup times, one of the areas it really shines in is creating small scripts for automating tasks, much like you would with shell-scripts. But it was lacking the basic stuff to query files, so I figured that was a nice little project, to add a fstat like function to Which, conceptually is quite easy, you run some stat like function in the Objective C part of the code, then expose that through a function in

##Here be dragons!

As a total Objective C n00b, I had to dig around in the docs to figure out what I needed to do. I ended up using NSFileManager:attributesOfItemAtPath:path:error which seemed to work ok. It returns a NSDictionary from which you pull the values, you create a javascript object, JSObjectSetProperty on it and Bob is your uncle. Unless he isn’t.

Because not all file attributes are present in the dictionary that NSFileManager returns. So you might want to test if they are present.

NSDictionary has a method [NSDictionary doesContain] which should do the job, but doesn’t. Probably because of some equals semantic, so you end up having to use [NSDictionary objectForKey]. Which has a really subtle thing about it. You see, when doing

if ([result objectForKey:NSFileAppendOnly] != nil) {
   bool appendOnly = [ result objectForKey:NSFileAppendOnly ];
   JSObjectSetProperty(ctx, retval, JSStringCreateWithUTF8CString("appendonly"),
                       JSValueMakeBoolean(ctx, appendOnly),
		       kJSPropertyAttributeReadOnly, nil);

the if test would actually return true even though the key was not present!

And the even more scary thing was that it would return false when stepping through the debugger! The solution to this was scary.

What’s the point of all this

The point of this is that this is quite similar to the code I write in my day-time job. The problem at hand is not very complex nor is it very difficult and it results in code which basically is sequential with little need for tests. Apart from the tests you need in order to ensure that you’ve understood the thing that your’re integrating with.

Here I’m integrating against OS X through an Objective C api, and my problems lie not in my complicated glue code, but rather in my understanding of the api.

And the tests you need to write to ensure you’ve understood the API very easily become mocks (if you know how to write those for the api you’re working against, I sure don’t know how to write mocks for Objective C) which only reflect your understanding of the API.

And one last thingy

There are of course several versions of OSX. Mike Fikes, the creator of Planck, apparantly has a host of macs running different versions of OS X. And while testing an early version of my patch, he found that the way I was formatting dates to ship over to Clojurescript didn’t work on OSX Lion, since my format-string yyyy-mm-dd'T'hh:mm:ssZZZZZ which returns "2013-08-25T00:16:27.000-00:00" on El Capitan, returns "2015-08-31T15:39:19GMT-04:00", the interesting bit being the added ‘GMT’, on Lion, which causes js/Date. to throw its hands in the air and give up.

So how do you write tests for this? I have no clue. My approach is to try and visually inspect the results until they’re good. I guess you could write a test which creates a file, runs js/PLANCK_FSTAT on it, and asserts that the output is somewhat expected, but that’s kind of offputting as well…