Slashdot Mirror
Taking On Software Liability - Again
An anonymous reader writes "You may remember an article in which a BBC correspondent wrote an article criticising current software licenses. In answer to the huge discussion that this brought about, he has written another article defending his views. From the article: 'It is possible to make error-free code, or at least to get a lot closer to it than we do at the moment, but it takes time and effort. Doing it will probably mean that commercially-available code is more expensive and cause major problems for free and open source software developers. But I still believe that the current situation is unsustainable, and that we should be working harder to improve the quality of the code out there.'"
I've said this years ago: software liability should apply on programs you pay for but for which you don't get the source. If money you pay goes to make something you don't have source level control over then that implies the vendor thinks its of sufficient quality that you, the end user, should not have to fix it. If you get the source then there is no guarantee and the distributor should have no liability. This doesn't mean you have to have the right to re-distribute the source -- but you have to have the right to re-build it using commonly available tools so liability can't be limited to one "magic" libarary.
[ ] vendor guarantees that software works as advertised
could be another checkbox that all software companies are trying to reach.
"What? You don't guarantee works-as-advertised? Well, then I'm looking for a different product."
If computing magazines would update their testing methods and added this one checkbox, Microsoft just might say "oh, hey, we haven't covered that checkbox yet. We need to have every checkbox. Let's quickly drop by the legal department get this in order..."
Dan Bernstein has offered a guarantee for many years that djbdns and qmail are secure. Now, this is a rather vague guarantee, since the task of deciding if a reported problem is a security flaw lies with Dan Bernstein himself; but it's a start.
I'm currently writing some cryptographic code, and I intend to go considerably further: I intend to offer a guarantee not only that my code operates as specified, but also that it is not vulnerable to any side channel attacks within certain classes.
As the time-to-exploit of security flaws continually decreases, I see only one solution: Writing code which is correct in the first place. If you can do that, you can offer a guarantee. And hopefully once security becomes as larger issue to consumers, people will start looking for guarantees.
Tarsnap: Online backups for the truly paranoid
Truly error free is not a likely state for software.
http://www.rootstrikers.org/
First off, I should issue a disclaimer that I'm an oldbie. I started programming in assembly language on punch cards, but no, this isn't going to be a rant about youngsters and their newfangled languages. (At least it better not be; my current job has me living, breathing, and eating PHP.)
The problem with bad software today -- just like it was thirty years ago -- is bad engineering. It's not because of the methodology du jour (or its absence), licensing, choice of language, or toolsets. You can write brilliant, bug-free, efficient software in COBOL using the basic procedural structured programming paradigm. You can write awful, buggy, resource-hungry software in object-oriented Java using XP. None of that shit matters.
Good engineering requires, among other things, a detailed understanding of the problem, thorough planning, the sheer experience required to distinguish between the clever and overcomplicated on one hand, and the lucid and elegant on the other, excellent communication between developers, foresight (also borne of experience), and rigorous debugging. All of these things, including the many other prerequisites not mentioned, require lots of time and effort. Too much time and effort, in fact, for most commercial software outfits to invest and still turn a profit.
That's the rub, really. All the methodology and language fads aside, the basic principles of good software engineering were worked out decades ago, and sometimes further -- good generic engineering practices in the abstract were worked out long before we harnessed electricity. It all comes down to this: the more time, effort, and care you put into a product, all other things being equal, the better the product will be. It's easy (and well-deserved) to mock Microsoft for the shoddiness of their major products, but that very shoddiness is why you can buy MS Word for less than ten grand. If MS built word processors the way engineers built the Golden Gate Bridge, the prices would be comparable.
The market does not reward that kind of quality. In the first place, no one is willing to pay thousands of dollars for a supremely excellent product when one that is good enough can be had for a couple hundred. Most folks couldn't afford that kind of software engineering even if they wanted it. In the second place, once you have the perfect all-in-one software package, why would you ever buy another one? Microsoft is in this position already with its good-enough products. No one needs an upgrade, so remaining profitable requires MS to churn out new versions of its increasingly resource-intensive operating system so that you at least have to buy new copies as you replace your older machines.
FOSS is at least theoretically invulnerable to these pressures. In theory, there will eventually be all-singing all-dancing FOSS packages covering all of the major software categories, and the age of commercial mass-market software will be at an end. I've been waiting for this day to come since well before the first release of Linux. I'm surprised that it hasn't come yet. I'm surprised that the majority of FOSS software is still as buggy, poorly designed, and -- almost without exception -- undocumented as its commercial equivalents.
I suppose I shouldn't be surprised. Excellence in software engineering is like excellence in any other field: it's really fucking hard. It's even harder when you have a day job; time constraints aside, after 8-12 hours coding at work, the last thing many developers want to look at when they get home is compiler output. Many of the remainder are either amateurs or students -- not to diss either category, but often the necessary experience is lacking, and the lone hacker often lacks the knowledge or the inclination to produce code that's easy for other developers to work with. I remain confident that we'll get there, though. (I am less confident that I will still care by then, but it will still be a boon to those who live to see that day.) I am equally certain, for the reasons
Proud member of the Weirdo-American community.
I thought about this the other day, asking myself why we can't have the same approach in software development as bridge building, or other engineering disciplines. The difference seems to be that of prototypes. When you build a bridge you create a prototype, test it as much as possible, tweak it where necessary and let the cycle continue until there is a working solution. Once that is done you are ready to build the bridge, based on specifications that in a certain sense are easier to follow than what software does.
Look at software and ask yourself where that prototype is, that can tweaked reworked until all obvious and so obvious issues have been tested for? You will end up noticing that the prototype and the final product is the same thing. While a bridge can be tested based on a number of complex mathematical formula, I am not so sure that software can be tested in the same way. Software is designed and developed based on a number of philosophies and sometimes these even have to interface with other programs based on other philosophies. Over time the complexity grows to a point where testing it 100% is like trying to predict what the stock market is going to do next week. I would like to give a figure to what we are able to predict, but that I will leave that for someone else, since I am not sure I am qualified to do so.
At the same time I will say that there are a good number of things for which you can create unit tests for and these help avoid the most obvious issues. The non-obvious issues, based on difficult to reproduce scenarios, variable dependencies are a little trickier.
Things are also improving thanks to libraries that implement much in the way of reusable code, but here too there is an issue. Imagine that you designed your program to be dependent on libraries x, y and z, and then the user adds libraries that effect the libraries you depend on, how can you predict what is going to happen?
You will notice that most mission critical systems are designed to have only the most essential features (as compared to desktop software) and are often coded with very precise memory management and sometimes even avoid the pointer type and instead using only primitives. Trying to develop most applications this way would be long and laborious and your users would be complaining that his complex office software doesn't do what (s)he wants (remember they can't agree on what they want), even if it is 99.999% stable.
I am not saying it is impossible, its just that I have yet to see an approach that is 100% effective and for 100% of cases. Yes I am a software developer, so I do have a certain bias.
Jumpstart the tartan drive.
Let's not forget "another piece of software's fault." Installing Software package B might overwrite a registry setting or DLL needed by software package A. On top of that, software package B might leave something running in the memory as a service that conflicts with something software package A does.
You are correct, there are WAY too many variables when dealing with software failures. And if this guy were actually a software developer he'd know that it's pretty much impossible to make something completely bug free. The most you can hope for is something that rarely has a bug or recovers if it encounters ones without losing its place/data.
You are a civil engineer.
I want you to build a bridge.
I won't say where- or what the end conditions are on each end- because this bridge needs to work in about 2 million different places.
Now- as to what will cross the bridge. I won't tell you that either. It might be a car- it might be a convoy of tanks.
Now... as to the basic laws of the universe (the operating system). I can't tell you much about them either. For example, gravity may change at any time to be higher or lower. The tensile strength of various materials may change unpredicatably with various patches to reality.
Your work force will be available to work 2 to 16 hour days and may or may not comprehend instructions written in english.
The bridge needs to be built from scratch from materials using new refining methods so you cannot use any reference materials to analyze how strong it has been historically.
Finally, this bridge must be made of at least 9 million different pieces (opcodes). The subunits will be assembled by a robot of some kind (Compiler) so you will not know the details of how the units work- only how they are supposed to work as units.
---
I'm sorry but you really do not understand what you are talking about.
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.