Afterwood

I was recently watching an episode of Marvel’s Agents of S.H.I.E.L.D with my youngest before his bedtime. In the episode they had travelled back in time to 1940s America and were changing into clothes considered more appropriate for the times, to blend in. After they all got changed and met back up again one character remarked to the other, “Nice threads!” In this instance they were referring to another person’s clothes, but it got me thinking that nobody has ever said that to me, which, given my weak fashion skills is not surprising, but more topically for this publication, it’s not a phrase I’ve ever heard somebody remark about another person’s code either. Given the problems that all too often arise from the introduction of additional threads into a program it’s far more likely that you’ll be chastised for your threads rather than commended for them.

The late Russel Winder was a programmer who was no stranger to problems involving concurrency and had the good fortune to work with some serious parallel hardware when some of us were still all gooey eyed over the second CPU in our desktop machine. I’m pretty sure Russel would never congratulate anyone on their choice of threads as he was a big proponent of solving concurrency problems by using higher-level concepts. Like so much in the world of Computer Science many of the techniques for managing concurrency have been around for decades and Russel was always keen to promote the Actor Model, Communicating Sequential Processes (CSP), etc. in his talks and writings. I never really grokked either of these until Russel published his Introduction to GPars in CVu 22(6) back in 2011. I’ve still never written a line of Groovy or done anything significant on the JVM but this article provided the clarity I needed to start seeing how these ideas were realized in a modern language.

If you grew up in the UK during the 1980s you might already have an aversion to threads due to the BBC’s film of the same name which depicted the state of Britain after a nuclear war. I was slightly too young to watch it first time around, although like any teenager that didn’t stop me trying to because apparently there were ‘other kids’ in our school who had allegedly watched it. Despite the grown-ups slapping a 15 certificate on it to advise us youngsters against being foolish and dabbling in issues we were emotionally under-equipped to deal with, we jumped right in anyway and regretted it later. Why do we never listen to our elders? As the old joke goes, “Some programmers, when confronted with a problem, think ‘I’ll use threads’, now problems two have they.”

I once interviewed someone for a highly technical role and casually asked them how they felt about lock-free programming. They simply replied, “I’ll give anything a go.” While I admire their positive outlook on life, this was not really the response I was expecting. Anthony Williams’ book on concurrency with C++ weighs in at six hundred pages and Joe Duffy’s concurrency book for Windows hits nine hundred. What this tells me is that it isn’t something you can ‘dabble in’. It feels more like a career in its own right.

Debugging multi-threaded programs is always an interesting prospect, especially trying to single step through functions which are executing similar workloads on different threads. It reminds me of my first foray into the Usenet way back at university before the common availability of ‘threaded newsreaders’. Contrary to what you might be thinking, these weren’t programs which used multiple threads to achieve better UI responsiveness (we’re talking Unix terminals here), this was about stitching together a continuous stream of forum posts on different topics so that you could focus on one conversation at a time instead of constantly context switching between subjects. Single stepping through a multi-threaded program is always a bit of a shot in the dark as you wonder how far you’ll get before you’re whisked away to another land. At least we eventually get to return to where we left off unlike poor old Sam Beckett in Quantum Leap. “Oh, boy!” indeed.

It wasn’t just late 80s newsreaders though – this was also how Twitter felt in its early years. Fortunately, everything was largely unrelated anyway so there wasn’t really a context to be dragged away from and return to. The introduction of the 280 character limit in late 2017 was swiftly followed by the appearance of ‘threads’ as Twitter tried to convince its users that brevity was no longer the soul of wit. Further correlation between social media platforms and concurrent programming are possible when you consider their problems with coherence and false sharing.

One career that seems to have died out since the C++ committee finally decided to come clean with C++11 and define a thread-aware memory model is that of answering Stack Overflow questions about how to safely implement a Double-Checked Lock (DCL). The Internet was awash with solutions on how to safely acquire a Singleton (though, once again, you now have two problems) that turned out to be wrong. The dominance at the time of the Intel CPU meant that ‘works on my machine’ was almost a statistically valid argument. A few people working on more advanced CPU architectures got bitten but the strong view taken by the incumbent x86 meant that many of us lived in ignorant bliss. When Herb Sutter declared to the world that the free lunch was over, he was talking about CPU single core performance, but he might as well have been talking about the rise of the ARM which has a weaker view on ordering and a stronger view on messing with your head. Java had its DCL crisis just after Y2K calmed down whereas .Net had another decade to go before its bubble finally burst, even the JVM & CLR are not immune it seems.

With the continued working from home and decline of the suit and tie in the workplace, I suspect my chances for a fashion compliment have long since passed. As for the prospect of never having to deal with a threading issue again, all I can say is, “Promises, promises!”

Chris Oldwood is a freelance programmer who started out as a bedroom coder in the 80s writing assembler on 8-bit micros. These days it’s enterprise grade technology from plush corporate offices the comfort of his breakfast bar. He has resumed commentating on the Godmanchester duck race but continues to be easily distracted.