Slashdot Mirror
Code Quality: Open Source vs. Proprietary
just_another_sean sends this followup to yesterday's discussion about the quality of open source code compared to proprietary code. Every year, Coverity scans large quantities of code and evaluates it for defects. They've just released their latest report, and the findings were good news for open source. From the article:
"The report details the analysis of 750 million lines of open source software code through the Coverity Scan service and commercial usage of the Coverity Development Testing Platform, the largest sample size that the report has studied to date. A few key points: Open source code quality surpasses proprietary code quality in C/C++ projects. Linux continues to be a benchmark for open source quality. C/C++ developers fixed more high-impact defects. Analysis found that developers contributing to open source Java projects are not fixing as many high-impact defects as developers contributing to open source C/C++ projects."
Sunlight is the best bleach.
Java project developers participating in the Scan service only fixed 13 percent of the identified resource leaks, whereas participating C/C++ developers fixed 46 percent. This could be caused in part by a false sense of security within the Java programming community, due to protections built into the language, such as garbage collection. However, garbage collection can be unpredictable and cannot address system resources so these projects are at risk.
This is especially amusing in light of all the self-righteous bashing that C was getting over OpenSSL's problems. Seems it's true that using a "safe "language just makes the programmer lazy.
First, we shouldn't confuse Coverity's numerical measurements with actual code quality, which is a much more nuanced property.
Second, this report can't compare open source to proprietary code, even on the narrow measure of Coverity defect counts. In the open source group, the cost of the tool is zero (skewing the sample versus the commercial world) and Coverity reserved the rights to reveal data. Would commercial customers behave differently if they were told Coverity might reveal to the world their Coverity-alleged-defect data?
Again, having good Coverity numbers can't be presumed to be causally related to quality. For example, Coverity failed to detect the "heartbleed" bug, demonstrating that the effect of bugs on quality is very nonlinear. 10 bugs is not always worse than 1 bug; it depends on what that one bug is.
The report doesn't really go into an important measure.
What is the defect density of the new code that is being added to these projects?
Large projects and old projects in particular will demonstrate good scores in polishing - cleaning out old defects that are present. The new code that is being injected into the project is really where we should be looking... Coverity has the capability to do this, but it doesn't seem to be reported.
Next year it would be very interesting to see the "New code defect density" as a separate metric - currently it is "all code defect density" which may not reflect if Open Source is *producing* better code. The report shows that the collection of *existing* code is getting better each year.
But that's written in C and C is the worst programming language ever!! How dare they be so dumb to not write Python in a "memory safe" langauge!
Coverity: Hey you, proprietary software developer with the deep pockets. Yeah, you. We've got this great tool for finding software defects. You should buy it.
Proprietary software developer: get lost.
Coverity: Hey, open source dudes, we've got this great defect scanner. Want to use it? Free of course!
Open source dudes: Meh, why not?
Coverity: Hey proprietary software developer, did we mention those dirty hippie neck beards are beating the stuffing out of you in defect (that we detect)-free code?
PSD: Fine, how much?
This is a useless analogy. Code Quality is a function of both skill and the stewardship of the team supporting the code. Tools help as well but you can write some elegant, high quality code regardless of the language chosen. You can also write some real shit too but ultimately how many defects a piece of software has comes down to the design and testing that goes along with it. Some bodies of work get rigorous testing and it's not like OpenSSL's recent problem wasn't about deficient design it was about a faulty implementation. Faulty implementations in logic happen all the time and there are some bugs that just take awhile to become known. I mean even with test driven development and tools for code analysis probably couldn't have found this particular issue but considering how long it was in the code base without somebody questioning it goes back to not only stewardship by the team but the rest of the world who are using the code. If anything this situation points out that FOSS can have vulnerabilities just like proprietary software however the advantage is that with FOSS you can get it fixed much more quickly and because other people can see the implementation it can become scrutinized by folks outside the team that develops and maintains it.
In the case of Heartbleed the system works. A problem was found, it was fixed it's now just a matter of rolling out the fix and regressions are put into place to help insure that it doesn't happen again. The repercussions of what it means is that another gaping hole in our privacy was closed and that "bad guys" may have stolen data, rollout the fix ASAP. Your guess is as good as mine as to what was stolen is a matter of research and conjecture at this point. I doubt that the bad guys will tell us what they gained by exploiting it. Let's also be sure that until the systems with the bug are patched, they're vulnerable so cleanup on aisle 5.
To be honest it's a bit naive if we all assume that FOSS software that handles security doesn't have potential vulnerabilities. Likewise it's also naive to assume that proprietary code has it licked as well given the revelations of NSA spying for the past year. Given that there are numerous nefarious companies that sell vulnerabilities to anybody who can pay for it, that means unless you're buying them you probably will never know what is exposed until somebody trips over it. What this means for everybody that you can depend on is when those vulnerability-selling companies are out of business can assume that your software is free of the easier to exploit vulnerabilities; governments will always use all their tools to get intelligence including subverting standards and paying off companies who can give them access to what they want.
Harrison's Postulate - "For every action there is an equal and opposite criticism"
I'm not going to yell about the openSSL guys.
I'm going to be honest here, they deserve yelling at, and I'm an open source fan. The error they made is exactly the same mistake that everyone else has made in years past when dealing with SSL: x509 and the SSL protocol demands [lengthofstring][string], "pascal" style. This is how everyone (open and closed source) got hit with that domain validation bug where the certificate said "(26)bank.com\0.blahblahblah.com". Certificate signers looked at the domain on the end of the string "blahblahblah.com" and validated it. Client programs treated it like a C string and thought it was a certificate for "bank.com". Not a single person anywhere said "whoa there, null bytes are not part of a valid hostname!"
The attack asks server to respond with "(65535)Hello" and the server replies with 65535 bytes of data. Falling for this attack is exactly like the guy who points and laughs at the person who just fell off their bike, seconds before falling off their own bike. They should have known better, especially with how high-profile these attacks were in the past.
The bit about writing their own malloc implementation, poorly, was just icing on the cake.
with nearly 2x the LOC.
"none, because someone might find out I've made the code worse, not better"
I would expect both "open source" code to be of approximately equal quality to proprietary code. In each ideology you will get people who care (about quality), and people who don't, in approximately equal proportions, the same with skill, ingenuity and passion for the work.
The difference is that proprietary software is constrained by the number of developers able to view and work on the code. An open source project may have a similar number, or smaller set of core developers, but a much larger pool of developers that can spot problems, suggest alternatives, fix the one bug that is affecting them, etc. Having a more diverse set of developers will increase the chances that the software improves.
You could also make an argument about the motivations of the developers. Open source projects are often a community of people passionate about what they are building and have a strong incentive to make their code readable by others. By the nature of open source a developers reputation is on the line with every bit of code they make public. I've met far more developers scared to make their horrible code public than those worried about getting fired for equivalently horrible code.
Complexity Happens
Disclaimer, I work for Coverity. There's a write-up on why Coverity didn't find it out of the box here:
http://security.coverity.com/b...
With all the noise about OpenSSL lately, running this Coverity test on it (and other security software like GNUTLS) and sharing the results seems like it would be a good thing...
Your four-sentence comment has five glaring errors that make it obvious that you have absolutely no idea what you're talking about. You very much remind me of the job applicant who told me he has experience in C, C+, and C++.
If you have good quality people, especially a good leader, your code will be good.
Even if the people are relatively inexperienced.
At this point, just about everything in IT/CS is a research project, not innovation.
So it's a matter of diligently doing the work based on past archetypes.
Futurist Traditionalism
Some open source projects will have better code then closed source projects and vice vesa, you can't just make a clean line.
are you sure about that?
that's valid C#, all you need to do is inject something like that into the codebase and let the JIT compile it (using all the lovely features they added to support dynamic code) and you're good to get all the memory you like.
Now I know the CLR will not let you do this so easily, but there's always a security vulnerability lying around waiting to be discovered that will, or an unpatched system that already has such a bug found in any of the .NET framework, for example this one that exploits... a "buffer allocation vulnerability", and is present in Silverlight.
The moral is ... don't think C programs are somehow insecure and managed languages are perfectly safe.
Yes, so your argument is that you can, with great difficulty cause a possible security issue in C#, but in order to do so, you have to basically say... I'm about to do something possibly bad, please don't check to make sure what I'm doing is bad. Then modify the compiler from default to allow said code to be compiled, then put it into a fully trusted assembly so it bypasses all security checks, and THEN you might have an issue.
and this is comparision to where in C/C++ where you can write an exploit in 2 lines of code by accident, using nothing but defaults.