Code quality, NIH, golden code, and module ratings

I'm pretty critical about code quality, be it for my own projects or those of others, and have become known as giving harsh code reviews for the Drupal community. Hearing "ask Morbus, but be prepared to cry" pleases me, and I have little impetus to stop doing things I enjoy, regardless of what people think. Recently, someone suggested the reason my reviews seem "assholish" is because I state "you" a lot. I agree with this, I think, and have made an attempt to lessen my usage, though not my anality.

I define code quality via a number of different hard-to-describe metrics. In the case of Drupal, following the code style guidelines is certainly key, but also following Drupal core patterns as well - doing things "like core" is pretty high up there. This could be as simple as using the same documentation string for Drupal hooks, using the same pattern for the null option of a select box, et cetera. Using coder.module (a Drupal lint) to correct 90% of your mistakes should be required and, sometime in the future, tests from something like SimpleTest should be too.

I also think that, if you're going to "take over" a module, or otherwise release one, it should make an attempt at meeting the highest level of code quality. One could argue that "release early, release often" dissuades us from this, but I disagree: there's a difference between "doing it right" and "continuing to suck". I have no problem with weekly releases if you make an attempt to maintain or improve code quality. Hearing "A lot of strings have been lingering in incorrect forms ... as I try to prevent translations from breaking" isn't a legitimate excuse: Drupal core doesn't care about backwards compatibility, and it highlights your own inability to spell-, grammar-, or read-aloud check your strings (while also calling into question the quality of translators who have converted your broken documentation without complaint). I didn't release "my" version of Case Tracker until I considered it of high quality, and I made sure to provide explanation for what I was doing.

Since I've stopped working on Case Tracker, the CHANGELOG.txt hasn't been updated but, more importantly, it hasn't been removed either. This is just the first of a dozen "heading back to crap" problems that Case Tracker is suffering under, as I watch the new maintainer commit style error after style error. I don't give two shits if you're a certain type of programmer, or if you've coded this way "for years": if you're committing a Drupal contribution, you should follow the goals and styles of the parent project. Anything else is utter failure. If you don't want to, get the hell out of contrib, because it's already a sodding mess and you'll only make it worse. Requiring developers to read even the most basic of code practice books should be a guideline for access into any repository. "What is the name of the checklist on page 579 of Code Complete First Edition?", and other "you must have the book" tests, would be awesome.

This lack of quality control is the same reason Steven Wittens (UnConeD) left. He and I have gotten into a number of arguments, certainly, and he coined "The Morbus Effect" but, ideally, we're after the same thing: the best the code can offer, and an attention to research, detail, and testing that is, at times, incredibly boring.

Which all leads us into the blanket response to any criticism of code: "Where's the patch?". The overuse of the "code is golden, talk is not" ideology is as frustrating as anything else. I have no problem with contributing patches to Drupal modules which have made an attempt at code quality. There's a difference in "your code is good, but this needs fixing" vs. "holy crap, everything about this module is wrong". If you make an attempt to write good code, I'll put forth the effort to make it even better. If you don't, then I'm not going to waste my time improving your code: it just needs to start over. Make a new file and port things carefully, rewriting and organizing them properly. Or, do it in the same file and separate the good stuff from the bad by gigantic swathes of whitespace. Until you make the effort to show you care about your code and the parent project it belongs to, I'm not going to either.

Quality is one of the prime reasons I wrote an (unreleased) Achievements module as opposed to installing User Points. Ignoring the different approaches of "achievements" vs. "accumulate[d] points" (which I'll probably touch on in a later post), the 2.0 version of User Points failed at nearly everything. Besides showing, to me, a lack of quality, it actually didn't work properly, further exasperated because it shipped with a number of modules that were never maintained. I sent an email or two stating my dislike, and marveled privately about how so much crap could make it into a "2.0" release, especially one that was proudly presented and ballyhooed at various Drupal conventions. This is why I don't mind complaining about it publicly - it positions itself as something great, but I've yet to see any indication of that. I don't think I'd be as aggressive if it were a new module or developer innocently asking for help - there's no marketing, ego, or pride to provoke my ire.

Inevitably, I got into a chat about it on IRC, and that ever-nefarious "patch?" lament appeared. To me, User Points 2.0 was an example of wrong from the ground up, with duplicate data in the database, a UI that didn't work (functionally or architecturally), shipped modules that no developer cared about anymore, and crazy code patterns that drove me up a wall. The same conversation promised a new developer had taken up the reins of 3.0 with lots of rewrites, and I should check it out. Hating duplicate functionality and the whole "Not Invented Here" problem that often plagues plugin-driven development, I told myself that I would.

I'm embarking on such a journey now, as I find a need for my Achievements module once again. And, still, as I look at the very first lines of User Points 3.0, I find myself clucking disapprovingly. I don't see much improvement but, certainly, it requires more effort to truly make the decision (to their credit, they did remove all the unmaintained contributed modules). For what it's worth, my first 15 minutes of discontent equate to:

  • Abuse of constants: User Points defines its permissions and other strings inside constants. This isn't only wrong (Drupal core doesn't do such a thing) but serves no discernible purpose besides being an additional layer of abstraction - perhaps the developer's IDE supports constant completion? If you're not manually and anally testing every line of code (which would root out misspellings or typos, one of a few rationales for code completion in an IDE), then your module isn't up to par. Gladly, SimpleTest integration in the future will indicate more obvious dedication to testing and non-laziness but, of course, that presumes properly written and maintained tests, which is probably too much to ask and could lull one into a false sense of complacency.
  • String retardation: Administration permissions in Drupal core are always in the form of "administer something", but User Points happily uses "admin userpoints". It also doesn't know what the hell it's called: the module's page has "User Points", but usage within the code indicate it could be "Userpoints" or "userpoints" and that it could integrate with "e-commerce" or "ecommerce". These are all things I consider easy and obvious indications of quality control - most module interaction is only ever to an end-user, and if you can't perfect the most visible output of your effort, then the actual code isn't getting any love either.

At a 15 minute glance, if someone yelled out "patch?", I'd decline. This is a module that is showing the same problems I lamented privately about for 2.0, and just needs to spend an entire release cycle worrying about quality not functionality. With that said, note that this is, literally, 15 minutes of glancing at the source. I've yet to actually install the new 3.0 and there's still a good chance I'll find it usable enough to merge with my achievements code. "Not Invented Here" syndrome is a sign of laziness too but, as jsled remarked after reading an early draft of this entry, "at some point, learning how to use a tool is more costly than just writing a special-purpose tool".

One suggestion for improving module visibility in Drupal is to offer module ratings. I wouldn't support this idea unless there were two distinct metrics: the end-user metric (does this module do what I want it to do?) and the developer metric (does this module's code live up to the code we ship in Drupal core?). To me, a module's quality equates to a combination of both, and any rating system that attempts to merge the two, or leaves one out, isn't adequately asserting overall health.

Tags: