Continued elsewhere

I've decided to abandon this blog in favor of a newer, more experimental hypertext form of writing. Come over and see the new place.

Tuesday, December 06, 2005

UML kills me

So I'm being forced to sit through a long and tedious course on UML, taught by a guy who probably hadn't started high school when I first began doing object-oriented programming. It's 80% tedium, 10% idiocy, and maybe 10% useful (if you accept that UML can actually be a useful communications tool, which I do).

So yeah, a standard visual notation for expressing software designs is a good thing. Nonetheless UML appears supremely braindamaged to me, for many reasons:

- the presentation of UML seems inextricably linked to a particular notion of software development process. Thus, you have diagram types that are sort of similar (concept and class diagrams) but whose main difference is that they are intended to be used at different phases of the process (analysis and design). Unnecessarily redundant, and they are close enough to each other to cause cognitive confusion.

- UML concepts are loose, vague, sloppy, and weirdly named. For instance, "attributes" and "associations". Both of these are (ultimately) links between an object and something else, but the first is used for primitive types and the latter for object-object relationships. Argh. But actually, when I quizzed the presenter about this, he said that associations turn into "association attributes" or maybe it was "Attribute associations". Double argh.

- UML is a visual language, but the tools used to create diagrams all suck like a black hole. OK, maybe they don't all suck, I haven't tried them all, but Visio and ArgoUML and Rational Rose do. They all have one or more of: lousy UIs, missing features, idiosyncratic notation (to pile on top of the built-in idiosyncracies of UML).

- Since the tools suck in different ways (for instance, ArgoUML is actually sort of OK but is missing some of the diagram types), it might be possible to compensate by using more than one and passing the UML between them. Ha. UML is a well-established industry standard, right? You'd think that as a standard, it might be possible to write out a UML diagram from one UML editor and read it in with another. You'd be wrong. There is NO established serialization format for UML. This I find amazing and hard to understand even taking UML on its own terms.

- Related, but getting more philosophical, UML constantly fudges the extent to which it is trying to be a formal language. On the one hand, we are just drawing pretty pictures so the marketing guys can see what we are doing. On the other hand, we introduce all sorts of details that suggest we are really writing code, like public/private indicators on attributes. And then everybody tries to make tools that can do "round trip engineering" and go back and forth from UML to code (I've never seen one of these work).

- There's some undefineable grating quality about all discourse surrounding UML. I think it's related to the above, its attempting to straddle the gap between actual code and something else. This something else might be "business" but if so it's expressed in some way I don't understand.

I guess I'm extra miffed because I actually like diagrams and visual programming, but it seems to be that we have a case of a poor standard sucking all the air out of the space of visual object oriented programming. Also, I completely agree with what I see as the main goal of UML, which is enhanced communication between engineers and customers/domain experts. Again, I just think this is an abysmal way to go about it. Object-oriented programming is supposed to do this in itself, without an extra layer of sloppily-defined cruft.

Maybe I'm just missing some sort of business-oriented DNA. I went through an experience some years ago, where I (and my cow-orkers) first encountered relational databases. Everybody there was from sheltered academic environments where we had never seen such a thing before, and we regarded mostly with hostility and didn't make very good use of it. Eventually I had an epiphany where I finally got the value of the relational model, and understood how real-world applications might actually be interested in data. Perhaps UML similarly has value that is just hidden from me due to my lack of appropriate background, although I doubt it.

Keyword: venting,
Google fodder: UML sucks

1 comment:

Anonymous said...

UML sucks indeed, dunno which retard invented it. I am checking out SysML. http://www.omgsysml.org/ I would love to have a modelling system which was Pythonic, if you know what I mean.