[jerf]

"This is an even more conservative position than the default Chesterton's Fence reading, which is itself dismissed by a lot of people as pedantically restrictive."

In a normal, real-life context, I can see why someone would feel that way.

In a software engineering context I think it's just a further emphasis that you ought to understand what something is doing before fiddling with it, and both the original intent and what it is currently doing are interesting information. Many times I've removed dead code, only to learn that not only was it alive (which wouldn't have been that surprising, it's easy to miss some little thing), but that it is very alive and actually a critical component of the system through some aspect I didn't even know existed, which means I badly screwed up my analysis.

The differences between the physical world and the software world greatly amplify the value of the Chesterton's Fence metric. In the physical world we can all use our eyes and clearly see what is going on, and while maybe there's still a hidden reason for the fence that doesn't leap to mind, it's still a 3-dimensional physical problem in the end. Software is so much more interconnected and has so much less regard for physical dimensions that it is much easier to miss relationships at a casual glance. Fortunately if we try, it's actually easier to create a complete understanding of what a given piece of code does, but it is something we have to try at... the "default view" we tend to end up with of software leaves a lot more places for things to hide on us. We must systematically examine the situation with intention. If you don't have time, desire, or ability to do that, the Chesterton's Fence heuristic is more important.

We're also more prone to drown under Chesterton's fences if we're not careful, though. I've been in code bases where everyone is terrified to touch everything because it seems like everything depends on everything and the slightest change breaks everything. We have to be careful not to overuse the heuristic too. Software engineering is hard.

[wpietri]

In software, I would strongly disagree with this:

> you ought to understand what something is doing before fiddling with it

I think understanding before fiddling is one option. But I think a better option is often fiddling and seeing what happens. The trick is to make it so that fiddling is safe. E.g., on a project where I have good test coverage and find mystery code, I can just delete it and see what fails. Or I set up a big smoke test that runs a bunch of input and compares outputs to see what changes.

A lot of bad software is effectively incoherent, so it can't be understood as machinery. Instead it has to be understood in psychological, historical, or archaeological terms. "Well back then they were trying to achieve X, and the programmer doing it was not very experienced and was really interested in trendy approach Y so he tried using library Z, but it wasn't really suited for the problem at hand, so he misused it pretty severely."

That can be interesting, but it's often much more efficient to say, "Who cares how or why this got to be such a tangled mess. Let's solve for the actual current needs."

[jefftk]

> But I think a better option is often fiddling and seeing what happens. The trick is to make it so that fiddling is safe. E.g., on a project where I have good test coverage and find mystery code, I can just delete it and see what fails. Or I set up a big smoke test that runs a bunch of input and compares outputs to see what changes.

There was a good discussion of this in the comments, starting at https://www.jefftk.com/p/accidentally-load-bearing#fb-109168...

In addition to tests, you can also add logging to your running system, or make the change behind a flag that you A/B test in production.

[bombcar]

The problem “fiddle and see if Rome burns” is that you may have broken something that only runs once a year/decade and you won’t know if that process isn’t in the list of tests.

And by the time it breaks, will anyone remember the probable cause?

[wpietri]

If people are building things that only run once a year or once a decade and haven't included sufficient tests to make sure those things work and keep working, then the failure lies with them, not with the people who work on the system later. The real "probable cause" isn't some later change. It's the initial negligent construction.

[jerf]

"But I think a better option is often fiddling and seeing what happens."

If you're safely fiddling with it, I would consider that part of the process of understanding. I'm particularly prioritizing understanding what is actually doing. Historical context is primarily useful because it will point you in the direction of what else you may need to look at; when I fail to realize that removing X broke Y it's because I didn't realize that there's a reason those are together.

[wpietri]

I agree, but what I'm talking about is distinct from a usual "process of understanding", because the goal is explicitly remain as ignorant as possible. In many cases I want to learn the exact minimum needed to safely make a change. What were they thinking? What motivated them? Don't know, don't care. I just want to clean up the mess and move on to something actually useful and pleasant.

As a practical example, last year I was supposed to figure out why a data collection system was not getting all the data we wanted. The person who worked on it was long gone, so I looked at the code. It was a bunch of stuff using Apache Beam. No tests, no docs. The original author was not a very experienced programmer. It jumbled together a variety of concerns. And after a day of trying to understand, it became obvious to me that some of the uses of Beam were unconventional. Plus Beam itself is its own little world.

The next day, I said "fuck that", and wrote a very simple Python requests-based collector. I pretty quickly had something that was both much simpler and much more effective at actually getting the data. And from there I went into the usual sort of feedback-driven discovery process to figure out what the current users actually needed, with zero regard for original intent of the system was in solving problems for people no longer present.

What was the thing doing? Why did it to it? How did it end up that way? All excellent mysteries that will remain mysteries, as I eventually deleted all the Beam-related code and removed its scheduled jobs. For me, this ignorance was truly bliss. And for others too in that the users got their needs solved with less work, and that developers had something much cleaner and clearer to work with in the future.

[marcosdumay]

> But I think a better option is often fiddling and seeing what happens.

Yeah, as always, IMMV.

But I do agree that online discourse puts too much emphasis on statically analyzing systems, and too little on adding instrumentation or just breaking it and seeing what happens.

At the same time, my experience is that on practice people put too much emphasis on instrumentation or just breaking it and seeing what happens, and way too little on statically analyzing the system.