My problem with this discussion is that a lot of people just say "I'm a scientist (or I'm working with scientists) and I'm observing X so I can say 'scientists blahblahblah'".

Different scientific research fields are using widely different computer software environment, and have their own habits and traditions. The way a biologist uses programming has no reason to be similar to the way an astrophysicist does: they have not at all experienced the same software environment. It may even be useless to talk about "scientist" in the same field as two different labs working in the same field may have very different approaches (but it's more difficult if there are shared framework).

So, I'm not at all surprised that you observe opposite experience. The same way I'm not surprised to see someone saying they had the opposite experience if someone says "European people are using a lot of 'g' and 'k' in their words" just because they observed what happened in Germany.

Absolutely my experience as well. Scientists write code that works, but is a pain to reproduce in any sort of scalable way. However it’s been getting better over time as programming is becoming a less niche skill.

The problem I've run into over and over with research code is fragility. We ran it on test A, but when we try test B nothing works and we have no idea why because god forbid there is any error handling, validation, or even just comprehensible function names.

I wish we communicated the intent of the “best practice” instead of the practice itself.

Actually, when I’ve followed those guidelines, it’s because the tech lead graduated in the 1980s, almost certainly learned it all on the job, but has always done it that way. Others just do what they’ve done before. School talked about those things, but not in a “this is the right way” sort of thing.

There is no best practice. It is good to know the tools. In dojo, do that crazy design pattern shit and do crazy one long function. Do some C#, Java, JS, Go, Typescript, Haskell, Ruby, Rust (not necessarily those but a big variety). I want the next person to understand my code - this is very important. Probably more important than time spent or performance. If spending another 10% refactoring to make it easier to understand, even if just adding good comments, it is well worth it. Make illegal state impossible, if you can (e.g. don't store the calculated value, and if you do then design it so it can't be wrong!). Make it robust. Pretend it'll page you at 2am if it breaks!

Such as what? I don't really know of any such superstitions that are based on nothing.

I see a lot of opinion/taste presented as something more, but I really can't think of superstitions.

It is important to have popular and powerful tools that can reduce amount of code for things like caching and building.

For example, Snakemake (os-independent make) with data version control based off of torrent (removing complication of having to pay for AWS, etc) for the caching of build steps, etc would be a HUGE win in the field. *No one has done it yet* (some have danced around the idea), but if done well and correctly, it could reduce the amount of code and pain in reproducing work by thousands of lines of code in some projects.

It's important for the default of a data version control to be either ipfs or torrent, because it's prohibitive to make everyone set up all these accounts and pay these storage companies to run some package. Ipfs, torrent, or some other centralized solution is the only real solution.

Today's "best practice" is tomorrow's worst practice.

Yes, the claim that the scientists' hacked-together code is well tested and even uses valgrind gave me pause. It's more likely there are no tests at all. They made a change, they saw that a linear graph became exponential, and they went bug hunting. But there's no way they have spotted every regression caused by every change.

Agree with those two problems on the scientist side. I would also add that they often don't use version control.

I think a single semester of learning the basics of software development best practices would save a lot of time and effort in the long term if it was included in physics/maths university courses.

1 and 2 are features. Re 1, if someone doesn’t know what the output should look like they shouldn’t be reusing the code. Re 2, just think a bit more about it and you’ll realize fretting over ram that isn’t needed until it’s needed is actually just premature optimization.

Right and I think "scientists" simply are more intelligent than average Joe Coder. Intelligent people produce better software.

It is easy to learn some coding, not so easy to become a scientist.

To becomes a scientist you must write and get your PhD-thesis approved, which must already be about scientific discoveries you have made while doing that thesis. Only people with above average IQ can accomplish something like that, I think.

Yeah somehow I suspect this author hadn't yet had to deal with colab notebooks.

Yeah, well gnome 2 also doesn't compile or run on my machine. It ran once at some point, but one is considered a "worse" class of software.

>They are also less productive when coding than the scientists because they care too much about the quality of their work and not enough about getting shit done.

Ultimately I’d say the core issue here is that research is complex and those environments are often resource strapped relative to other environments. As such this idea of “getting shit done” takes priority over everything. To some degree it’s not that much different than startup business environments that favor shipping features over writing maintainable and well (or even partially) documented code.

The difference in research that many fail to grasp is that the code is often as ephemeral as the specific exploratory path of research it’s tied to. Sometimes software in research is more general purpose but more often it’s tightly coupled to a new idea deep seated in some theory in some fashion. Just as exploration paths into the unknown are rapidly explored and often discarded, much of the work around them is as well, including software.

When you combine that understanding with an already resource strapped environment, it shouldn’t be surprising at all that much work done around the science, be it some physical apparatus or something virtual like code is duct taped together and barely functional. To some degree that’s by design, it’s choosing where you focus your limited resources which is to explore and test and idea.

Software very rarely is the end goal, just like in business. The exception with business is that if the software is viewed as a long term asset more time is spent trying to reduce long term costs. In research and science if something is very successful and becomes mature enough that it’s expected to remain around for awhile, more mature code bases often emerge. Even then there’s not a lot of money out there to create that stuff, but it does happen, but only after it’s proven to be worth the time investment.

>I've found the problems that biologists cause are mostly 1. Not understanding dependencies, public/private, SCM or versioning, making their own code uninstallable after a few months

That's not on them though. That's on the state of the tooling in the industry.

Most of the time, dependencies could just be a folder you delete, and that's that (node_modules isn't very far from that). Instead it's a nightmare - and not for any good reason, except historical baggage.

The biologists writing scientific programs don't want "shared libraries" and other such BS. But the tooling often doesn't give them the option.

And the higher level abstractions like conda and pip and poetry and whatever, are just patches on top of a broken low level model.

None of those should be needed for isolated environments, only for dependency installation and update. Isolated environments should just come for free based on lower level implementation.

I was going to write a response but you've put what I would have said perfectly. The problem, at least in academia, is the pressure to publish. There is very little incentive to write maintainable code and finalise a project to be something accessible to an end user. The goal is to come up with something new, publish and move on or develop the idea further. This alone is not enough reason not to partake in practices such as unit tests, containerisation and versatile code but most academic code is written by temporary "employees". PhD's a in a department for 3-4 years, Post Doc's are there about the same amount of time.

For someone to shake these bad practices, they need to fight an uphill battle and ultimately sacrifice their research time so that others will have an easier time understanding and using their codes. Another battle that people trying to write "good" code would need to fight is that a lot of academics aren't interested in programming and see coding as simply as means to an end to solve a specific problem.

Also, another bad practice few bad practices to add to the list:

* Not writing documentation.

* Copying, cutting, pasting and commenting out lines of code in lieu of version control.

* Not understanding the programming language their using and spending time solving problems that the language has a built in solution for.

This is at least based on my own experience as a PhD student in numerical methods working with Engineers, Physicists, Biologists and Mathematicians.

These patterns appear in many fields. I take it as a sign that the tooling in the field is underdeveloped.

This leads to a split between domain problem solvers, who are driven to solve the field's actual problems at all costs (including unreliable code that produces false results) and software engineers, who keep things tidy but are too risk-averse to attempt any real problems.

I encourage folks with interests in both software and an area of application to look at what Hadley Wickham did for tabular data analysis and think about what it would look like to do that for your field.

I work in an R&D environment with a lot of people from scientific backgrounds who have picked up some programming but aren't software people at heart. I couldn't agree more with your assessment, and I say that without any disrespect to their competence. (Though, perhaps with some frustration for having to deal with bad code!)

As ever, the best work comes when you're able to have a tight collaboration between a domain expert and a maintainability-minded person. This requires humility from both: the expert must see that writing good software is valuable and not an afterthought, and the developer must appreciate that the expert knows more about what's relevant or important than them.

> They are also less productive when coding than the scientists because they care too much about the quality of their work and not enough about getting shit done.

You can't solve the first 3 issues without having people who care about software quality. People not caring about the quality of the software is what caused those initial 3 problems in the first place.

I only worked briefly in software for research, and what you described matched my experience, but with a couple of caveats.

Firstly, a lot of the programs people were writing were messy, but didn't need to last longer than their current research project. They didn't necessarily need to be maintained long-term, and therefore the mess was often a reasonable trade-off for speed.

Secondly, almost none of the software people had any experience writing code in any industry outside of research. Many of them were quite good programmers, and there were a lot of "hacker" types who would fiddle with stuff in their spare time, but in terms of actual engineering, they had almost no experience. There were a lot of people who were just reciting the best practice rules they'd learned from blog posts, without really having the experience to know where the advice was coming from, or how best to apply it.

The result was often too much focus on easy-to-fix, visible, but ultimately low-impact changes, and a lot of difficulty in looking at the bigger picture issues.

> Not understanding dependencies, public/private, SCM or versioning, making their own code uninstallable after a few months

I'm not sure what "uninstallable" code is, but why does it matter? Do scientists really need to know about dependencies when they need the same 3 libraries over and over? Pandas, numpy, Apache arrow, maybe OpenCV. Install them and keep them updated. Maybe let the IT guys worry about dependencies if it needs more complexity than that.

> Writing completely unreadable code, even to themselves, making it impossible to maintain. This means they always restart from zero, and projects grow into folders of a hundred individual scripts with no order, depending on files that no longer exists

This is actually kind of a benefit. Instead of following sunk cost and trying to address tech debt on years-old code, you can just toss a 200-liner script out of the window along with its tech debt, presumably because the research it was written for is already complete.

> Foregoing any kind of testing or quality control, making real and nasty bugs rampant.

Scientific code only needs to transform data. If it's written in a way that does that (e.g. uses the right function calls and returns a sensible data array) then it succeeded in its goal.

> They are also less productive when coding than the scientists because they care too much about the quality of their work and not enough about getting shit done.

Sooo...another argument in favor of the way scientists write code then? Isn't "getting shit done" kind of the point?

Yeah these problems with "engineer code" the author describes, they are real, but it's a well known thing in software engineering. It's exactly what you can expect from junior developers trying to do their best. More experienced programmers have gone through the suffering of having to work on such code, like the author himself, and don't do these mistakes. Meanwhile, experienced scientists still write terrible code...

I'm a software engineer working with scientist-turned-programmers, and what I've experienced is also exactly the opposite of the author. The code written by the physicists, geoscientists and data scientists I work with often suffers from the following issues:

* "Big ball of mud" design [0]: No thought given to how the software should be architected or what the entities that comprise the design space of the problem are and how they fit together. The symptoms of this lack of thinking are obvious: multi-thousand-line swiss-army-knife functions, blocks of code repeated in dozens of places with minor variations, and a total lack of composability of any components. This kind of software design (or lack of design, really) ends up causing a serious hit to productivity because it's often useless outside of the narrow problem it was written to solve and because it's exceedingly hard to maintain or add new features to.

* Lack of tests: some of this is that the scientist-turned-programmer doesn't want to "waste time" writing tests, but more often it's that they don't know _how_ to write good tests. Or they have designed the code in such a way (see above) that it's really hard to test. In any case--unsurprisingly--their code tends to be buggy.

* Lack of familiarity with common data structures and algorithms: this often results in overly-complicated brute-force solutions to problems being used when they needn't have and in sub-par performance.

This quote from the author stood out to me:

> I claim to have repented, mostly. I try rather hard to keep things boringly simple.

...because it's really odd to me. Writing code that is as simple as it can be is precisely what good programmers do! But in order to get to the simplest possible solution to a non-trivial problem you need to think hard about the design of the code and ensure that the abstractions you implement are the right ones for the problem space. Following the "unix philosophy" of building small, simple components that each do one thing well but are highly composable is undoubtedly the more "boringly simple" approach in terms of the final result, but it's a harder to do (in the sense that it may take more though and more experience) than diving into the problem without thinking and cranking out a big ball of mud. Similarly reaching for the correct data structure or algorithm often results in a massively simpler solution to your problem, but you have to know about it or be willing to research the problem a bit to find it.

The author did at least try to support his thesis with examples of "bad things software engineers do", but a lot of them seem like things that--in almost every organization I've worked at in the last ten years--would definitely be looked down on/would not pass code review. Or are things ("A forest of near-identical names along the lines of DriverController, ControllerManager, DriverManager, ManagerController, controlDriver") that are narrowly tailored to a specific language at a specific window in time.

> they care too much about the quality of their work and not enough about getting shit done.

I think the appearance of "I'm just getting shit done" is often a superficial one, because it doesn't factor in the real costs: other scientists and engineers can't use their solutions because they're not designed in a way that makes them work in any other setting than the narrow one they were solving for. Or other scientists and engineers have trouble using the person's solutions because they are hard to understand and badly-documented. Or other scientists and engineers spend time going back and fixing the person's solutions later because they are buggy or slow. The mindset of "let's just get shit done and crank this out as fast as we can" might be fine in a research setting where, once you've solved the problem, you can abandon it and move on to the next thing. But in a commercial setting (i.e. at a company that builds and maintains software critical for the organization to function) this mindset often starts to impose greater and greater maintenance costs over time.

[0] https://en.wikipedia.org/wiki/Anti-pattern#Big_ball_of_mud

This was my exact experience working in biomedical hpc.

    >    * Not understanding dependencies, public/private, SCM or versioning, making their own code uninstallable after a few months

This is definitely true, but I've searched *far and wide* , and unfortunately it's not a simple task to get this right.

Ultimately, if there were a simple way to get data in the correct state in an os-independent, machine independent (from raspberry pi to HPC the code should always work), concise, and idempotent way - people would use it. There isn't. But the certainly could be.

The solution we desperately need is a basically a pull request to a simple build tool (make, Snakemake, just, task, etc) that makes this idempotent and os-independent setup simple. Snakemake works on windows and Unix, so that's a decent start.

One big point is matching data outputs to source code and input state. *Allowing ipfs or torrent backends to Snakemake can solve this problem.*

The idea would be to simply wrap `input/output: "/my/file/here"` in `ipfs()`, wherein this would silently check if the file is locally cached to return, but if not go to IPFS as a secondary location to check for the file, then if the file isn't at either place, calculate it with the run command specified in Snakemake. It's useful to have this type of decentralized cache, because it's extremely common to run commands that may take several months on a supercomputer that give files that may only be a few MBs (exchange correlation functional) or only a few GBs (NN weights) so downloading the file is *immensely* cheaper to do than re-running the code - and the output is specified by the input source code (hence git commit hash maps to data hash).

The reason IPFS or torrent is the answer here is for several reasons: 1) The data location is specied by the hash of the content - which can be used to make a hash map of git commit hashes of source code state that map to data outputs (the code uniquely specifies the data in almost all cases, and input data can be included for the very rare cases it doesn't) 2) The availability and speed of download scales with popularity. Right now, were at the mercy of centralized storage systems, wherein the download rate can be however low they want it to be. However, LLM NN weights on IPFS can be downloaded very fast when millions of people *and* many centralized storage providers have the file hosted. 3) The data is far more robust to disappearing. Almost all scientific data output links point to nothing (MAG, sra/geomdb - the examples are endless). This is for many reasons such as academics moving and the storage location no longer being funded, accounts being moved, or simply the don't have enough storage space for emails on their personal Google drive and they delete the database files from their research. However, these are often downloaded many times by others in the field - the data exists somewhere - so it just needs to be accessible by decentralizing the data and allowing the community to download the file from the entire community which has it.

One of the important aspects to include in this buildtool would be to ensure that, every time someone downloads a certain file (specified by the git commit hash - data hash map) or uploads a file after computing it, they host the file as well. This way the community grows automatically by having a very low resource and extremely secure IPFS daemon host all of the important data files for different projects.

Having this all achieved by the addition of just 6 characters in a Snakemake file might actually solve this problem for the scientific / data science community, as it would be the standard and hard to mess up.

The next issue to solve would be popularize a standard way to get a package to work on all available cores/gpu/resources, etc on a raspberry pi to HPC without any changes or special considerations. Pyspark almost does this, but there's still more config than desirable for the community, and the requirement of installing OS-level dependencies (Java stuff) to work on python can often halt it's use completely (if the package using pyspark is a dependency of a dependency of a dependency, wet lab biologists [the real target users] *will not* figure out how to fix that problem if it doesn't "just work"[TM])

Putting some effort into (3) can increase your citations (h-index). If people can’t use your software then they will just find some other method to benchmark against or build on.

Here you are not improving your time to get out an article, but reducing it for others - which will make your work more influential.

> 3) Time spent on documentation, optimization or anything else that does not directly lead to results is directly detrimental to your progress.

Here's is where I disagree. It's detrimental in the short term, but to ensure reproducibility and development speed in the future you need to follow best practices. Good science requires good engineering practices.

Congrats, you used design patterns.