I’ve been a heavy user of the D programming language since 2013. The appealing thing about D both then and now is its potential as a glue language. Back in the old days, we’d talk about scripting languages as glue languages, with R and Tcl being two of the more prominent examples. You can create an R package containing D code and use it from R like any other package you installed from CRAN. You can have R, Python, D, and C code in the same program. D provides all the low-level power of C with a nice, intuitive syntax. The learning curve is low compared to C++ or Rust.
Over the years, I’ve invested a lot of my leisure time in D. A couple projects I shared publicly were embedrv2 and BetterR. The former allows you to write libraries of D functions and handles all the boilerplate so that you can call them from R without doing any additional work. The latter is a very comprehensive solution for data science in D. I’ve done a lot more, but a lot of the time that went into these particular projects was the boring stuff to make it accessible for others. This is powerful stuff. D is very much a suitable choice for data analysis.
Time to Embrace Reality. I haven’t wanted to admit this, but it’s clear that there is no market for D as a language for data analysis. I’m not talking about the usual use cases for R, Python, or Stata, where you’re calling a canned routine as part of an interactive data analysis. There’s no market for D for applications that require speed. As far as I can tell, nobody has used any of the work I’ve shared.
For this reason, effective immediately, I’m abandoning my efforts to promote D as a language for data analysis. Although my projects will remain active, and I will continue to use them for myself, they shift to “here’s the code, have fun with it” status. There will be no more time spent writing documentation, creating good examples, getting stuff to work with Dub, or figuring out how to get anything to run on Windows. As before, others are welcome to contribute any of these things. I’ll be happy to work with them, but I won’t be the one doing the work. Not that I expect anyone to contribute.
Changes in the Programming Landscape. I should also disclose that I wouldn’t recommend D for data analysis. Things have changed dramatically since 2013. Today, if an economics grad student asked me to recommend a fast language, I’d tell them to use Julia. Someone wanting a traditional compiled language should look at Swift - though I haven’t done enough with it to recommend it. It’s been used for machine learning applications. Importantly, it’s a heavily-used language with significant resources backing it.
D is Not Perfect. As D’s advantages have faded over time, due new languages like Swift and improvement in older languages, its weaknesses for data analysis are harder to ignore. One of the big limitations is operator overloading. Anything numerical requires working extensively with operators in order to be usable. Old-style Java or C-style function calling is a non-starter. Without any operators, x = 3*4+7 would be x = plus(times(3, 4), 7). The ugliness and verbosity, as bad as it might be, is less of a problem than the difficulty of writing correct code. The only real solution is to break it out into small pieces and then put all the pieces together. I understand the argument against operator overloading. It’s just that what is a good idea in some domains is a non-starter for data science. Now, D does have some operator overloading, so the situation isn’t as bad as it could be, but they’re real. Not being able to represent matrix multiplication and element-by-element multiplication is a serious limitation. The inability to overload the comparison operators is a catastrophe. The people arguing for these restrictions are doing so without an understanding of how data analysis gets done in languages like R.
Other limitations of D include package management and working with DLLs on Windows. Look at package management in R and Julia and then think about the silliness of telling users to write a JSON configuration file and put their files in the right directories. Problems with Windows is not unique to D. The compiler could nonetheless be used to lessen the burden.
Direction of the Language. I hesitated to write about this, but I’ve lost all confidence in the direction of the language. Here’s one example. A library author complained about the need to support so many compilers. There’s a standard solution to that in the open source world - you put a label like “LTS” or “stable” on a release and then library authors only test against those releases. Never satisfied with simple, proven solutions to simple problems, the D community decided to go with a spectacularly overengineered solution called “editions”. Sixteen months have passed, library authors still have to deal with a large number of compiler releases, and I can’t tell you about editions because they’re still putting it together. It would not surprise me at all if this remained an issue in 2028.
The one that’s most severe for me is the planned move to safe by default. A language that’s been developed for more than 20 years with C interoperability as its biggest selling point isn’t the best candidate for safe by default. It’s obvious that this is a situation where you do something like add a compilation flag for safe mode. Nope. That’s not happening for unstated reasons. The plan is to impose safety on everyone and to make it a pain in the ass to use unsafe code. It’s hard for me to comprehend the decision process that leads us to these places. That would mean that in order for someone to use D for data science (which requires interoperability with other languages) they’d need to deal with that. Yikes.
Unmoderated Forum. One example that might not seem like a big deal to others is the lack of moderation on the D forums. I can’t tell a student to post there if I don’t know what someone else is going to post in response. Even if the responses are okay, it’s not a good look to point to what is often a cesspool as representative of the language - it simply is not professional.
What All This Means. I’m not saying D is dead or anything like that. It’s a good language, has a significant user base, and I’ll keep using it for my own purposes. I’m no longer going to promote it to others. I’ll tell them to use other languages and use other languages for my teaching.