One of the worst things you can do in software development is assume things. You see it a lot - developers assume that they know why software is slow, so they don't profile. They assume they know what the problem is, so they don't test. They assume they know what the requirements are, so they don't ask.
I fell into that trap on the bug side over the last 3 releases of BottomFeeder. Starting with version 3.7, I integrated Software With Style's XHTML viewer (and more recently, their editor) in the application. The viewer doesn't handle general HTML - it really wants XML (which means XHTML for general web content). Now, most stuff out there just isn't XHTML. In fact, a lot of the stuff out there is really bad, because browsers have been built to be lenient. What to do?
Well, the first thing we did was use the parser in Twoflower. That worked ok, but - as it happens (and I knew this, having used Twoflower for all the previous releases of BottomFeeder), it doesn't handle every possible breakage in HTML. There was plenty of content that the Tf parser wasn't cleaning up well enough to view. So, we looked at LibTidy - an existing C library that cleans up HTML quite nicely.
Michael built a Smalltalk wrapper for it - Windows and x86 Linux - and I started shipping it with BottomFeeder. That's where I got into trouble. It's been years (over a decade now) since I worked regularly in C. All of the knowledge I once had has rusted away, and I'm now very reluctant to delve down to the level. So, when we got occasional crashes in BottomFeeder, I made an assumption - I decided to add an option to the application that allowed the user to toggle libtidy on/off, and chalked the error up to general C level instability.
Well. As you can see from this post, the problem was much, much simpler, and involved a stupid mistake in the Smalltalk wrapper code - one that popped out at me as soon as I sat down and looked seriously at the code. So by making an assumption - rather than actually looking at the code - I cost myself 2 releases of a less stable application. That's the kind of problem that you can buy yourself by not actually looking.