Report: Aging Java Components To Blame For Massively Buggy Open-Source Software

itwbennett writes: The problem isn't new, but a report released Tuesday by Sonatype, the company that manages one of the largest repositories of open-source Java components, sheds some light on poor inventory practices that are all-too-common in software development. To wit: 'Sonatype has determined that over 6 percent of the download requests from the Central Repository in 2014 were for component versions that included known vulnerabilities and the company's review of over 1,500 applications showed that by the time they were developed and released each of them had an average of 24 severe or critical flaws inherited from their components.'

Comments

The root cause : poor unit testing

[Dr_Barnowl]

Why?

Because if you don't test your code, you don't know if changes to it break it.

Changing the components your code is composed of is a big change.

Therefore : people get nervous about changing the components they have used (even changing the version).

What should be happening : when you're planning a new release, raise the component versions to the latest and run your test suite. If it passes, good job, release it.

What is actually happening : the version numbers never get edited, because that version worked, and if you change it, OMG, it might stop working.

Re:The root cause : poor unit testing

[DickBreath]

The root cause: poor management (in most cases)

The root cause is not poor unit testing. Not bad developers. It is managers who won't allow the change to be made. It ultimately will always come down to money. They are unwilling to spend on having a reasonable staging environment that closely mimics the production system such that making these changes could be done safely and receive proper testing. And people to do that work also cost money.

In short: management doesn't care, due to money. So the product can just self-destruct. (like SourceForge)

Re:The root cause : poor unit testing

[Yaztromo]

What should be happening : when you're planning a new release, raise the component versions to the latest and run your test suite. If it passes, good job, release it.

What is actually happening : the version numbers never get edited, because that version worked, and if you change it, OMG, it might stop working.

Part of the problem I run into with this is that sometimes projects stick with old dependencies because at some point, some major version came along that significantly changed the organization of the API in such a way that the latest component version an't just be dropped in, but requires significant resources refactoring your code to use it. Getting management buy-in for that when there aren't any big customers breathing down their neck to get a flaw fixed can be neigh on impossible.

I ran into this recently myself. During internal testing, I discovered a flaw in our product when accessing any of our web resources using an IPv6 destination IP in the URL (i.e.: http://18080./ A quick bit of debugging showed that an external library we had been using for several years was doing some brain-dead parsing of the URL to pull out the port number; it was just doing a string split after the first colon it found, and presumed the rest was the port number.

Modifying the Maven POM to use a newer version of the API in question was initially difficult because the project had since reorganized their own library structure, breaking things into multiple smaller JARs. Except that some of the functionality was actually _removed_, and isn't available at the latest API revision (functionality we had been using, naturally). Classes had moved around to different packages than where they were previously, and various interfaces appear to have been completely rewritten.

Upgrading to a version of the library that actually fixed the flaw was going to be akin to opening Pandora's Box. Unfortunately, our former architect (from whom I inherited this code) was the type of guy who just liked to throw external libraries at every problem. In the end we had to document the fault for all current versions of the product, and now I'm trying to get management buy-in to do the work necessary to upgrade the library in question for the next version of our product. And this is for just one library out of over 100 that need similar attention.

Suffice to say, I'm not happy about this state of affairs. Unlike the previous architect, I push against using third-party libraries as our solution to everything. If I were allowed to rewrite everything from scratch, we could avoid these problems. Things are unfortunately messy out here in the real world, and when libraries decide to significantly change their interfaces your program uses to access their functionality, no amount of unit tests is going to make upgrading those libraries any easier.

Yaz

Re:The root cause : poor unit testing

[RabidReindeer]

This is somewhat deceptive. Sonatype supports Maven component archives.

One of Maven's chief claims to fame is that when you build a project, it doesn't grab "the latest" versions of dependencies, it grabs the selected versions of dependencies. On the grounds that "If it ain't broke, don't fix it".

This ensures a predictable product because everyone who does a build, no matter when, no matter where, will be pulling in the same resources to build with.

The problem arises when one (or more) of those selected component versions turns out to have issues. The build ensures that the product will be consistent, and thus will pass its own tests, but as the old observation goes, testing cannot prove the absence of bugs, only their presence. So if there was a vulnerability, an old project's tests wouldn't see it. And because you're asking for a specific library release version, later fixes don't get automatically included (of course, neither do later breakages, but they ignored that aspect).

In theory, then, this is simple to fix. Just update the project (POM) to pull in newer, better dependencies.

And the NEXT version of Windows will fix all your problems, and I've got a very nice bridge in NYC for sale cheap.

If you're working on a project, you generally have all you can do to keep up with issues in your own code, let alone some supposedly trustworthy third-party libraries. You cannot afford to be constantly updating the dependency versions and even if you could, there's the issue of "dependency Hell", where changing the version of Hibernate can conflict with the version of slf4j which can conflict with junit, which can conflict with... I usually like to budget 2 or 3 DAYS when I'm ready to start upgrading dependencies.

Sonatype doesn't get a pass here, though. If they/Maven supported a mechanism that could flag builds that have known weak dependencies, it would help a lot. Management, of course, would promptly command it to be turned off to ensure "productivity", but at least we'd have some help short of periodically manually auditing every library in a complex project (like that's ever going to happen).

Is this unique to Java?

[gstoddart]

I'm betting if you have a large enough pool of open source things, which depend on other open source things, then the bugs in the dependencies will trickle up to the projects which rely on them.

Though, admittedly, Java has also made this more annoying -- a decade or so ago when I was actively working on a Java project, it always amazed me how a new version of Java could completely break everything and then you'd have to re-test and re-certify everything.

It got to the point we put in very large bold characters in our release notes ... we work on this version of Java, if you get clever and introduce your own version of Java, we won't talk to you until you confirm the bug in the version we support.

A surprising number of clients were willing to blaze trail with whatever version of Java came along, and then kept expecting we'd be supporting custom versions from vendors or features which didn't exist when our version was built.

Eventually we learned to dread a new release of Java. Because invariably things went to hell and stopped working.

Re:Is this unique to Java?

[Anonymous Coward]

The last major release that "broke" things for me was the 1.4.2 -> 5 transition in 2004. Since then (5 -> 6, 6 -> 7, and 7 -> 8) have been relatively painless, If you were relying on an undocumented feature, or compiling against com.sun.* or sun.* classes you did so at your own risk. If you stuck to the documented JDK, you were usually ok.

Re:Is this unique to Java?

[Daniel Hoffmann]

Yeah this is a common problem in pretty much all platforms, what makes Java stand out is too many Java things are actually specifications, not implementations. It kinda mixes all the headaches of conventional development (dll hell, outdated libraries, testing against multiple hardware/OSs) with the headaches that developing for browsers have (multiple implementations of the same specifications). One of the things that make people like the Spring is that, unlike J2EE, Spring is an implementation, not a specification so it usually works is you change your application server for example. Well some parts of Spring relies on Servlets, which is a specification, but Servlets implementations are ok (although Java6 does not support Servlets 3.0 which is a pain in the ass if your client is on Java6 and refuses to update).

This is not surprising

[ErichTheRed]

This basically defines some of the problems of "enterprisey" software:
- It's composed of a million glued-together libraries.
- It's written by chronically understaffed/overworked IT department employees.
- Rigorous testing either (a) doesn't exist, (b) is so onerous that most developers try to avoid it, or (c) is outsourced/offshored to the lowest bidder, and therefore isn't completed without the staff basically doing the tests for the outsourcer.
- Anything that breaks it is avoided at all costs because of all of the above.

By extension, this is why some companies are stuck running IE 6 for key applications, or Office 97 because rewriting the scary mess of macros that runs a process isn't something anyone wants to do. I do systems integration work, and new versions of Java, web browsers, etc. are miserable. They introduce bugs small enough to be annoyances (rendering problems, etc.) and big enough to break the entire system.

The key to fixing this is for the software architects to require that developers move up to at least a semi-modern release of their key libraries, test everything against them, and remove the old outdated ones once all the bugs are fixed. The problem is that this is never done.

Re:This is not surprising

[JustNiz]

>> The problem is that this is never done.

The reason is that many Software Director positions are now filled with technically clueless people that are basically salesmen rather than engineers.
They have no comprehension of the concept of technical debt, or the need to spend time on activities that don't directly translate into new features.
The net result is that you're always just piling more crap onto the top of a steaming turd pile so making it worse, instead of working to replace the shit.

Accept the fact

[cloud.pt]

It's about time everyone stops whining. There are things in life you're better safe than sorry, but then there are things in life you just can't change: not every single entity can keep maintaining what they create. Human beings are limited, and so are human organizations - they lack money, workforce or simply the patience to put up with some "critical flaws" that are just too rooted in bad design to be solved without a restructuring.

THAT IS THE REAL FLAW.

There are good ways and bad ways to create reusable components. Black boxing (containing) everything for starters (sans the closed-sourced part) is something people tend to limit the scope to testing and/or to services outside a fully-fledged system's component border. Technologies like SOA are just one of many ways to plug&play every new piece of technology that performs a very specific task in a different way of a previously flawed one. Think project Ara. It's not only fun to develop like this (although some have problem conceptualizing it), but it's also more robust in the long run. Using such paradigms is what we, as the "clients" of such "aging and flawed" components can do push better development of individual components.

Now, each and every component developer has to find ways to keep their work atomic, so as to not conflict with the principles of technologies they are developed to work for. This might all seem like an utopian way of what to expect of the coding community, but then again we are also still looking for the best ways to apply near-perfect political views designed hundreds of years ago, which are yet to achieve full potential. I keep my hopes up for both issues, but my expectations low.

No mention of Sonatype's business?

[Captain Damnit]

It should be noted that the company releasing this report, Sonatype, markets a product called Insight Application Health Check that scans your binaries for libraries with known vulnerabilities.

I have never used their service, and can offer no comments on its utility or value. However, it is a bit unseemly that TFA doesn't mention that the source of their information about this very real problem also sells a service that solves it. This is a knock on IT World, not Sonatype.

Blame Maven

[_xeno_]

This is a problem that Maven has created, mostly.

What the summary doesn't mention is that "large repository of open source software" is a Maven repository. Maven allows you to specify dependencies for your Java project.

The problem is that you have to specify a specific version of whatever you use. So let's say you use OpenFoo 1.1 and that at the time you write your code, the latest version of OpenFoo is 1.1.3.

Now assume a horrible vulnerability is discovered in OpenFoo 1.1.3, so they release OpenFoo 1.1.4 to fix it. Well, your Maven POM says you require OpenFoo 1.1.3, so until you go in and manually change that, you will only ever use 1.1.3. There is - by design - no way to say "I want the latest 1.1 version." You can only describe a single, specific version.

So it's no surprise that Sonatype will see a ton of old Maven projects continuing to download outdated Maven artifacts. There's no way to say "I want the latest version of a specific branch" you can only specify a single version. Which means that a project that hasn't changed in years will still pull in the old versions of the libraries, even if it would work with the later versions.

Re:Blame Maven

[Anonymous Coward]

There's no way to say "I want the latest version of a specific branch" you can only specify a single version. Which means that a project that hasn't changed in years will still pull in the old versions of the libraries, even if it would work with the later versions.

No, Maven does support version ranges: You can say stuff like this: <version>[1.0.0,2.0.0)</version>. Here's a pretty good thread on the subject from StackOverflow.

To be fair, I don't think that most projects do this, but at least it's supported. Also, I'd guess that an analysis of a lot of the other artifact repositories like PyPi, Bower, or npm would produce similar results.

Re:FUD

[DickBreath]

It's not FUD. It's clickbait. It will make up for the revenue shortfall from SourceForge.

Sonatype FUDs Open Source ..

[nickweller]

April 2013: "Sonatype's annual survey of 3,500 software developers and shows struggle in setting corporate policy on open source and enforcing it" ref

April 2013: "Control and security of corporate open source projects proves difficult | New Sonatype survey finds 80 percent of most Java applications comes from open source" ref

Nov 2014: "Software developers use a large number of open-source components, often oblivious to the security risks they introduce or the vulnerabilities that are later discovered in them." ref

April 2015: "open-source also represents a vast, unpatched quagmire of cyber-risk that’s putting public safety at grave risk. That’s the assessment of Joshua Corman, CTO at Sonatype" ref