Language Features and Culture
Language Features and Culture
Why is Maven Central so full of libraries, and what kinds of community culture caused it?
Join the DZone community and get the full member experience.Join For Free
How do you break a Monolith into Microservices at Scale? This ebook shows strategies and techniques for building scalable and resilient microservices.
I was reading an interesting article on DZone recently that pointed out how many different programming languages have a rich community producing modules that are available for reuse. In some cases this community is not strictly affiliated with the creators of the language itself.
It's instructive to see just how many examples there are out there. These are just a few that popped into my head without thinking too hard.
- Perl has the Comprehensive Perl Archive Network (CPAN).
- R has the Comprehensive R Archive Network (CRAN).
- Python has the Python Package Index.
- Node.js has the Node Package Manager.
- Ruby has RubyGems.
- Java has Maven Central.
Together these repositories represent millions of programmer hours spent putting together libraries, and probably millions of hours spent maintaining the index and the servers on which they run. On the other hand, even though C and its semi-successor C++ have been around since before the Internet age, there isn't an exactly equivalent library and module system available.
Of course, C is the language of Linux, and so to some extent your favorite package manager is a C and C++ library and module system, but there's pretty clearly a difference in scope between what's available for C or C++ and what's available even for a relative newcomer of a language like Go or a language with a relatively small user base like Haskell. Compare Hackage with this list of C++ libraries.
Why is that? There's clearly a large and high quality user base for C and C++. I think that the issue comes down to language features and to one language feature in particular: garbage collection. To explain why I'll give an example.
As a result, I spent a lot of time in the Mozilla code base for someone who wasn't part of the team. I learned a lot about C++ from that code base; it remains a very impressive example of large-scale C++ development.
As expected, a decent amount of the core work that became the Mozilla codebase was centered around making the same C++ work well on Windows, Linux, and Mac, via two isolation layers: the Netscape Portable Runtime (NSPR) and the Cross-Platform Component Object Model (XPCOM). NSPR allows most of the Mozilla codebase to be written independent of the underlying operating system. It handles the kinds of things you would expect: threads, input / output, networking, etc.
XPCOM provides a number of cross-platform features similar to Microsoft COM. The most visible in terms of its impact in the Mozilla codebase is memory management in the form of "nsCOMPtr" smart pointers, similar to COM smart pointers. Basically, a smart pointer is a pointer that keeps track of how many things are pointing to it, and cleans up the thing it's pointing to when no one is using it anymore. It's an elegant alternative to garbage collection; instead of having a core runtime function that finds unused allocated memory and cleans it up, each new allocation of memory comes with a little bit of garbage collection intelligence.
Of course, in order for this smart pointer to work, any time a new reference to the pointer is stored outside a single function scope, the smart pointer needs to be notified so it can increment its count. And whenever that reference goes out of scope, the pointer also needs to be notified. And the consequences of missing one of these is pretty significant, either in the form of memory leaks or in the form of dereferencing a pointer that is no longer valid.
This is one of the most sophisticated memory management schemes available in a non-garbage-collected runtime. But it's hard to imagine how it could be made to work across a wide range of libraries. Either each library would have to build on top of a single implementation of smart pointers, or each library would have its own similar scheme to learn (and bridge between when integrating multiple libraries).
And that's exactly what we see when we look at some example libraries. Imagine an XML or JSON parser in C that provides data binding to a struct. The parser needs to allocate the memory for the struct (or array of structs), along with variable sized data inside. How long should this allocated memory stick around? The application code that uses the struct data ideally should not have to worry about the fact that it came from JSON, but somehow there has to be a way to identify that the application is "done" with the data. To solve this issue, the very well written json-c library provides its own reference counting that is very similar to COM smart pointers. Unfortunately, that leaves the onus on the using application to help out with reference counting, even if the corresponding data is passed to some other library that might use it for an unknown amount of time. The same kind of functionality can be found in GLib as used in the GNOME Toolkit, and in Boost.
All of these implementations are different, so a programmer using these three libraries would need to learn and remember three different APIs for handling pointer references. Under these circumstances, it's little wonder that libraries in C and C++ tend to be limited in scope and mostly serve to isolate from hardware or provide low-level implementation of a standard.
I for one consider this a very interesting thought. The primary argument in adding garbage collection to a language is that it raises the level of abstraction, allowing a clearer focus on the business logic of the application by avoiding boilerplate memory management code (and also thereby decreasing the chances of errors caused by memory management done poorly). But the broader effect is to create a different kind of culture where a large number of community-contributed libraries are possible. It neatly illustrates the way that the best features of a system are often emergent.
Opinions expressed by DZone contributors are their own.