Sunday, April 21, 2013

Nerd Food: C++ times, they are a changin' - Part I

Nerd Food: C++ times, they are a changin' - Part I

I think most C++ developers would agree that there has never been a better time to be a C++ developer than the present: after all, the C++ 11 standard was released adding a slew of modern features to the language and we are now seeing clang and gcc nearing 100% feature completion. In addition, there is an amount of activity around the language and libraries that I don't ever recall seeing in my professional career; the C++ 14 papers give a pretty good idea of the diversity and reach of this work.

In effect, C++ is negotiating the transition from a closed, vendor driven platform to an open, community driven project. These two articles will attempt to narrate the process, from the perspective of a long time C++ user.

From the beginning, Stroustrup had always believed that without libraries C++ would be a pretty useless language:

Without a good library, most interesting tasks are hard to do in C++; but given a good library, almost any task can be made easy.

Plain C++ was a pretty useless language: many developers reinvented the wheel, and libraries of poor quality were endlessly rewritten. Its hard to remember now, but in the nineties it was a rite of passage to write your own containers such as doubly-linked lists. Slashdot user VortexCortex gives a pretty insightful picture of those long forgotten days:

Well, I wrote my own Hash Map implementation. Before that I had my own Linked Lists too. Before C++ came along I even maintained my own OOP implementation in C with a spiffy pre-processor to convert syntactic sugar into the equivalent ugly object oriented C code riddled with namespace prefixes, indecipherable pointer math for dynamic (virtual) functions (actions), and statically complied variable (property) look-up tables for run-time type inspection (reflection).

It led to incompatibilities between codebases – My Entities+C wouldn't be compatible with your C+OOP. Hell, we didn't even use the same terminology, or necessarily even support the same feature sets. The point is, I wasn't the only one who was doing that (see also: Objective C). There were lots of us, it was a huge waste of duplicated effort. So many, in fact, that C++ was born, and the STL too […].

It is in this context that the C++ 98 standard emerged, and it was seen as a major victory for all involved - a milestone in the development of the language. However, pragmatism won the day and parsimony is always a requirement for shipping; the standard library was pretty much just the STL, the I/O Library and naively-defined string support. Once one exhausted the core functionality of the standard library, well, you were on your own. This is were library vendors came in.

C++ had always been a highly profitable tooling market for vendors, and the vacuum of specification only served to amplify the profits. It was now possible to make a true claim for vendor independence - "hey we are 100% standards compliant!" - while at the same time providing large amounts of vendor specific code. Once a vendor was chosen, vendor lock-in was almost by design: the libraries provided so much functionality and in such vendor-specific ways that changing libraries was synonymous to rewriting entire code bases.

The other interesting aspect of these days was that C++ community just didn't know how to write good C++. Most of the vendor libraries coped out and simply defined object oriented APIs because that was easier to create and easier to consume. Stepanov's dreams - that the STL was simply one example of many of a different way of thinking, and a raft of libraries along the same lines were soon to follow - were put on hold.

But people were already learning from the success of community driven languages such as Perl and Python. A man that was ahead of the curve in this regard was Beman Dawes. Not only did he deeply understand the standardisation process - and its flaws - but he also saw that libraries were the key to success. Thus Boost began as an effort to provide the missing libraries, in a nimble, community driven way that at the same time had high quality standards; these were provided by peer reviews, requests for documentation and so on. Boost soon became the home for the "fundamental" C++ thinking, in many ways continuing the discovery process of the Stepanov ideas in libraries such as Boost.Graph.

In the mean time, the world didn't stand still. The rich pickings C++ provided to vendors proved to be their downfall, as the largess of the tools market gave way to lean times. The crowds moved over to the new shinny toys such as Java and C#, and the new breed of dynamic languages gave C++ the final kick into the legacy bucket. It was just not cool anymore to be seen doing C++. These new languages learned their lessons and came out of the box with large class libraries that provided support for all sorts of things, making the standard library look small and antiquated in comparison. Even with Boost, one would still be nowhere near the functionality of out-of-the-box Java or C#, let alone Python or Ruby.

When a COBOL-style death appeared certain, a phoenix-like effect kicked in. Just as it was with UNIX - where the death and shrinkage of UNIX vendors became the catalyst to create a new breed of UNIX based on collaboration rather than competition - so the same happened to C++. When the big money moved away it was suddenly noticed that no single company could afford to maintain a large library by themselves on the skimpy margins they were now making. Thus many companies opted for open sourcing their code - some just before they went out of business, others as a strategic way of staying in business. Many vendors died, but the few that remained found healthy niches from which to run their businesses.

The second factor that kept C++ alive after all the cool kids moved on was that the old school engineers were still around, and when efficiency became a key concern - as it always does in the end - they were there to explain how efficient server side code could be written in C++. So whilst uncool, C++ was alive and well, just surviving underground in the guts of data centres and in large scientific projects. These users were making hefty use of open source, with Linux GCC and a plethora of open source libraries as their weapons of choice. Together with the appearance of Clang, this mean that both the tooling and the libraries in C++ moved forward in leaps and bounds, but in a process that was almost entirely divorced from the standard. From not having enough libraries we went to having too many of them, with lots of overlapping functionality.

Whilst this was happening, C++ didn't stand still. Some of the more obvious Boost code got shipped into TRs and made their way to end users. And C++ 0x was cooking. This was a major change to the standard, composed both of core language changes but much more importantly, of a great deal of additions to the standard library. A number of Boost libraries were polished up and moved into the std namespace.

But this brought with itself its own problems. Some were related to the new language features - its still not certain how some of these should be used - and others are related to the new libraries. For those of us heavily reliant on Boost, a difficult problem emerged: the standard libraries didn't cover all aspects of the Boost libraries they standardised; for instance, Boost Serialisation won't work with std::shared_ptr, so now code bases become a mix of Boost and standard library code. Clearly the process of offloading Boost libraries to the standard has only begun. And this is what we will be covering in part II.

Date: 2013-04-21 19:02:43 BST

Org version 7.8.02 with Emacs version 23

Validate XHTML 1.0

No comments: