|Perl: the Markov chain saw|
Re: Enterprise development: Its ok to say No!by Abigail-II (Bishop)
|on Dec 16, 2003 at 13:17 UTC||Need Help??|
But I've come to the conclusion that its only the profi that I want to be and work with.
That's a short sighted view, and isn't acceptable in the long run. Topping the 'profi programmer' is the programmer who can shape shift from one class of programmer to another, and knows when to be what kind of programmer.
Anecdote: I work for a company that has written software since the early 70's (I've coworkers that have been with the company or its predecessor for 30 years). I work in a department that is actively maintaining software that was written in the 70's (lots of FORTRAN and Pascal code) - part of the software is an OS that was initially written in PDP assembler, was later emulated on top of VMS, SCO, and is now running on top of Linux and Windows; where it runs as a secondary OS. We know all about portability, extensibility and maintainability.
We recently had a problem. As part of the nightly back procedure, the secondary OS calls a Linux program. This program does some mounting and umounting of disks hanging off fibre-channel devices and then queries /proc to get an impression of what disk are available. Due to a bug in the fibre channel driver, once in a blue moon, this generates a kernel oops, causing the kernel to kill the program, which in turn causes the secondary OS to dump and restart itself. For a particular customer, this was reason to not take a new release into production.
There were several ways of approaching this problem. The 'profi' programmer approach would be to dive into the driver sources try to find the problem, and fix the driver. The problem is, you can't estimate how long it's going to take, as it requires to become familiar with the driver. The 'happy customer' programmer approach is to write a wrapper around the program that causes the kernel oops, which just restarts the original program if causes an oops.
In the end, we did both. I wrote a shell script wrapper, and we send that to the customer the next day. Customer happy. Later, we ripped a newer version of the driver out of a newer kernel and implanted it in the kernel we're working with, and luckely that fixed the problem.
What I am trying to say is that the bigger picture is important. Maintainability, portability, extensibility, it's all important. But in the end, what's really important is (happy) customers. If your customers aren't happy - they leave. If they leave, your product can be fantastic, you won't survive. A good programmer balances the customers wishes and the development group wishes into an ideal blend.