I don't think things happen quite like that - what I have seen happen is that a "weekend kludge job" becomes the foundation for a lot of subsequent work and the "kludge job" sets the so after a year you can have a vital production system that has been extended and "improved" in a completely ad-hoc fashion and it's that that takes a lot of effort to reverse engineer and structure properly.

In my experience it is really difficult to raise the quality level on a project once it gets going - thing generally only get worse as external things start to impact (timescales, scope creep etc.). The only way things seem to keep high quality is to consciously start high and fight to keep standards high - which can be pretty difficult.

Like real life monetary debt, technical debt has 2 aspects: the principle and the interest. Sometimes the interest ends up being variable rate and backbreaking for your project - you should have written those tests up front and saved yourself a lot of time. Sometimes the whole project dies before you pay back the debt - you go bankrupt but it's no big deal, your creditors are friendly. In either case, the principle usually isn't smaller when you go to address it. You're just delaying payment in favor of speed.

What the original artical addresses are several aspects that I might call "technical assets" that have depreciated significantly. That kludge, if it wasn't done intentionally, isn't debt - it's now an asset. If you didn't have a plan to pay it off in the future, you've already bought it and now you need to replace it. And the replacement is very expensive. That is a technical investment you're deciding to make. You aren't just paying off debt at that point, it's a whole new capital expense.

To understand why, let's consider a normal interaction with a corporate legal team. The content team submits some copy to the legal team for approval and the legal team needs to communicate to the content team that they can't make the claim they are trying to make about the product. Now, it's more true in the technical sense for legal to say things like "Section 3 paragraph 7 of the Fair Sales Act states that Consumer Entities as defined in Section 1 Paragraph 3 may not represent their products in such a way as to potentially mislead a Purchaser as defined in Section 2 paragaph 9. Furthermore this issue has been thoroughly litigated at both the federal appellate and Supreme Court levels which have reinforced those interpretations unfavorable to the aforementioned Consumer Entity." But that wouldn't be great for either department. It's more useful for Legal to just say "You can't claim that - we'll get sued." In a similar way, it's more useful as a software engineer to just say "that decision will add to our technical debt" than to try and discuss the minutiae of how a certain bad architecture decision will make the system "resistant to change". As a fellow engineer, I'd rather have the latter conversation with you and the one about "technical debt" with my business manager.

"Technical debt" is the justification for process and development decisions that require more time up-front but will save time in the long run (especially as a term to communicate this to non-technical management). It only comes into play when you're talking relatively about two or more different choices.

Just as an example: when deciding to move ahead with less testing or to move more slowly and write unit tests for every function, you should consider the first to come with more technical debt. That doesn't mean it's the wrong decision, it just means that it should be a factor in your thought process.

If you look at a practice like Unit testing, you tend to find some organizations are having success with it and others not. Some of them see it as a "waste of time" and others see it as a "timesaver". How the practices are implemented are as important as the practices you decide to use. If the unit tests for your project take 3 minutes to run you may have a problem. If you are writing a large number of hard to maintain mock objects you may have a problem. If the program is not divided into units that are easy to test that is a different problem.

Another good example is the use of Vagrant. It is very possible to have Vagrant spinning like a top, but in a lot of places you hear software management grumbling that a team of 3 people wasted 2 weeks screwing around with it.

I see it so common for developer discussions to center around "cargo cult" positions (i.e. "singletons are evil") and not around a real dialog about the work to be done and how to do it.

They state, in general, it costs $3.61 technical debt / line of code.

"Technical debt is a way to understand the cost of code quality and the impacts of architectural issues. For IT to help drive business innovation, managing technical debt is a necessity. Legacy systems can constrain growth because they may not scale; because they may not be extensible into new scenarios like mobile or analytics; or because underlying performance and reliability issues may put the business at risk." (Tech Trends 2014, Deloitte University Press).

It's an easy to use metric to weigh the cost of supporting programs, and relatively simple for managers to understand. It has the added benefit of encouraging reducing the size of the code-base when possible.

An efficient algorithm may be punished over an inefficient one. An optimal one-liner can certainly cost a lot to develop (maybe you need a smart engineer, a lot of testing and a lot of time to figure it out) yet it would appear to be cheap.

Then there's just textual differences. It depends on the programming language. And it can be different even in the same language...

This function has one line:

    void f() { ... }

    void
    f()
    {
       ...
    }

    /*
    <several lines of comments>
    */
    void
    f()
    {
       ...
    }

The thing about "lines of code" is that it's CHEAP and EASY. It is also WAY better than having nothing. Which system is more complex, system A written in C++ 2003 or system B which was written Java in 2013? Which system is more complex, System C which is 200,000 lines or system D which is 10,000 lines? Even if I don't tell you how "lines" were defined in those two estimates, you can still tell more about C and D than about A and B.

"Lines of code" is a terrible metric, which can give only order of magnitude estimates. But that makes it enormously better than no metric at all, or a complex metric which we haven't actually measured.

A shorter function is a great side effect but it should never be the point. I want someone who can do things like: find obsolete code, create a better algorithm, decides on a better language for the task. If I say that a system is bad because it's 200,000 lines, I may discover that developers are really good at removing comments and obfuscating code to "improve" it down to 150,000 lines. If I come up with specific performance improvements, and state clear expectations such as "find functions we never use", I end up with a more maintainable system.

Let me be clear: I will defend "lines of code" as a metric useful for getting and order-of-magnitude estimate of the complexity of any given codebase. Anyone using it as a metric to determine how much to reward developers is just plain stupid.

[1] - http://www.folklore.org/StoryView.py?story=Negative_2000_Lin...

[2] - http://dilbert.com/strip/1995-11-13

Ideally, you want minimal technical debt. Any code you develop should aim for minimizing technical debt. Technical debt is not about evaluating the value of a line of code or program, but about it's potential maintenance cost.

I work at a large organization that maintains over 300+ modules of code and custom software programs, varying in size from a thousand lines of code to massive ones. Any one of which can be sold at any time, and all of which is supported in some fashion or another, and all of which should be able to work together.

You need some sort of metric to see where to invest ones time. If you don't manage the technical debt, everyone spends all their time doing tech support.

So, how the helpful concept of technical debt is evaluating where to put effort into improving the code base, and evaluating potential support cost of adding more modules? Lines of code, while flawed, provides a pretty good starting point.

If resources are handed out based on how bloated a project is, you would quickly discover how creative developers can be in increasing the relative "importance" of their code.

Systems are composed of lots of things, and sometimes a tiny piece is the most complex and critical. A measure of lines of code should not prevent you from adding 3 more people to a tiny project, and removing a person from a team with a giant code base.

https://www.youtube.com/watch?v=pqeJFYwnkjE

One of the other problems with the technical debt idea is, debt relative to what? Since I think most people do not consciously instantiate an answer to that question, I think most of us non-purposefully choose by default a comparison to an idealized perfect code base that is somehow perfectly correctly factored, yet also as fast as completely optimized code, completely documented without being overdocumented, simultaneously optimized for all possible future changes, and despite not existing at the moment, also perfectly understood by the original team such that they can do anything they like to it, and such understanding will survive any such refactoring, all accomplished in exactly the same amount of time that it took to write the terribad code that we actually have.

OK, but that was never on the table, though. What I described isn't even possible to manifest, and even if you back it down to what is at least sort of possible, you don't have the resources to manifest it before you simply run out due to lack of customers. You're not really taking on "debt" if you fail to manifest that, any more than you are taking on "debt" if you fail to make every possible dollar you could if you perfectly correctly harnessed your skills and made the perfect deals.

I think what a lot of people call "debt" is simply the natural state of code. Debt is something somewhat more exceptional than that. There is a core idea of value there, because there definitely is an operation where we deliberately make a short-term choice at the cost of long-term pain, but it requires more thought than I think has traditionally been given to it. The baseline needs to be more carefully defined.

As I said at the top, I'm just musing here; I don't have a ready-to-go definition of "baseline quality" here.

I don't think that's such a problem, though. The "perfect" codebase should be the baseline to which you compare your current debt.

Just like in real life, you take on some debt to have a basic standard of living. You might have a mortgage or maybe a car loan or some minor credit card debt, and you're normally always working on paying a little bit of that off. In a perfect world you'd want no debt at all, just like in a perfect codebase. But, a small amount is natural. It's only when it gets unruly and unmanageable that it becomes a problem.

---

Aside, the first and last lines of your comment reminded of this post: https://news.ycombinator.com/item?id=10859383

What I described isn't perfect; it's impossible. It destroys the utility of the idea of technical debt if that's the baseline, because it means that all choices are between "really bad" and "really bad".

"Aside, the first and last lines of your comment reminded of this post:"

Oh, certainly. But people do tend to assume that all comments that are not completely complimentary are intrinsically disagreement, and, well, let's be honest... statistically, it's true, so it's hard to be too annoyed that people's brains make that inference by default.

As it stands, over 5% of the codebase has been deleted so far without reducing functionality. One could argue that a significant part of the application were just hacks to get around the original hack, and hacks to get around those etc.

Not sure how to break that down into the categories provided, but given that the problems add up to proportions unimaginable to anybody but IRS, debt as an analogy works well enough for me.

As a business, if I buy a computer on a company credit card, I know immediately that I owe $1200.00 at 23 percent interest per year. In contrast, if I develop a piece of software and decide to use global variables to speed up development, I may immediately gain a mythical man-month in development time BUT I have no idea how much it may slow me down in the future if at all. Sometimes global variables are the way to go and development will be consistently faster for the entire lifetime of the software compared to using local variables. This tends to happen, not surprisingly, when the data in the global variables needs to be used widely throughout the program.

Avoiding effort now causes more effort to be spent later. The amounts are unknown but the effort is real.

Unfortunately management often thinks that they're supportive but ask them what project they need to axe or put on the back burner to free up time and the ephemeral promises and deferment to some hazy future date get put on the table.

Perhaps it's not "real" in development but it is very much a real issue that needs to be thought about by anyone in charge of operations.

You don't want a culture that only adds features, forever growing the pile. Developers need to be free to identify and aggressively throw out old functions, old tests, obsolete documentation or anything else. This means that a list of deprecated items should be expected in most projects, and mean it when you warn other teams that you plan to completely remove those items in 6 months.

This freedom to delete doesn't mean that total rewrites will be encouraged. Rather, it acknowledges that in an evolving system, some things do become obsolete and mistakes will be found (despite highly competent architects) that ought to be corrected. Ultimately it frees people to easily identify and maintain only what really matters.