Subject: Re: engineering counts
From: "Jonathan S. Shapiro" <shap@eros-os.org>
Date: Sun, 21 Oct 2001 13:51:56 -0400

> > The overhead and delay introduced when following documented
> > processes is huge for a small gain in quality
>
> I'm sure many of us have direct experience to the contrary.  The GCC
> group is mostly trending in good directions, for example.

I beg to differ. The work on EGCS is a great example of a group that has
consistently screwed up by failing to make any reasonable attempt at
regression testing. The EROS project had to freeze on the pre-EGCS line of
compilers because EGCS didn't work, plain and simple. Even RedHat had to do
this (kgcc). Things appear to be getting slowly better, but the recent
history is not something I would tout proudly.

As to whether the cost of documentation is worth it, I think there are
several issues here:

First, documentation is often done stupidly. We need to distinguish between
the true cost of documentation and the cost of stupidity.

Second, the actual measurements show that documentation done right
consistently results in a net reduction of end-to-end development cost.

Third, I think you are getting caught in the "throwaway" trap. If we have
learned anything in software it should be that nothing is throwaway.

Finally, I think you underestimate the impact of collateral damage from poor
software. My expectation is that software liability is about to hit us all
between the eyes because of this. This is a value chain that is outside the
bounds of the one normally adressed by marketing.


Jonathan