8 ms·
I suspect rbanffy is talking about things like avionics, medical device control, and the space shuttle. This is a whole other ball game, and it is possible to
by nostrademons 17y ago
I suspect rbanffy is talking about things like avionics, medical device control, and the space shuttle. This is a whole other ball game, and it is possible to ship software that is pretty darn close to perfect in those domains. They cost a few orders of magnitude more per requirement, and the requirements are sharply limited. These systems usually operate on a very limited number of input values and internal states, so you can often simply enumerate all the possible inputs and all the possible states to verify that the system works as designed.
- rbanffy 17y agoI wasn't thinking about these markets, but their mere existence proves perfect software can be written - it's possible. What I was thinking was the bold statement that doing something right (or perfectly right) is impossible and its widespread use as an excuse for not even trying. A software developer must aim for perfect software, even if shipping it is not always economically viable. The fact end-users have been beaten into submission and now tolerate having to restart their computers after software crashes or to redo calculations because the results were absurd the first time is not relevant to this discussion. If they tolerate less than correctness, we shouldn't. The current security problems we have are entirely fault of people who regard immediate profit as more important than long-term quality. And I take exception with anyone who suggests this is how things should be.
- tptacek 17y agoDan Bernstein didn't "regard immediate profit as more important than long-term quality". Neither did Wietse Venema. Both were at extreme pains to make sure their free software was free of security problems. Unlike any commercial software, security software included, software security was the #1 priority for both these projects. Both failed to ship code free of game-over security flaws. I find your point about "tolerating" failure to be irrelevant to the real world at best. At worst, it's just fatuous. I've shipped hundreds of thousands of lines of code, and I've led projects to review millions of lines more: chipset to JVM, embedded to desktop, open source and proprietary. It is a point of pride with us: we always find things. You think you're being respectful of users when you chant "don't tolerate failure". What you're really doing is being disrespectful of computer science. All evidence available to us suggests that it is practically impossible to ship code --- or at least, real-world code --- without security flaws. Is secure software an impossible goal? Not for computer scientists. For instance, a Stanford team build a system for evaluating code for flaws based on abstract analysis. That project became Coverity, a commercial product for scanning source code for flaws. They compete with a larger company called Fortify, in a hundred million dollar market for static code analysis. Projects like Stanford Checker, or the OS and runtime hardening work being done in countless labs, have a chance of advancing the state of the art to where you think it really is now. A line developer on a browser project has no such chance.