Rebooting Ourselves

It was a fairly rough week. Thus, over this weekend I thought about re-examining and resetting various procedures and things I do, as opposed to actively filling my time with activities. This reminded me of a Humans of New York post I came across several years ago:

“I’m rebooting my life entirely, again. It’s time for Andrew 5.0.”

In computer science, semantic versioning is a system for identifying different versions of software products in a standardised way. Under this system, a product’s version is an ordered triple of nonnegative integers written major.minor.patch. This is typically used for software, though the definition does not seem to require it. The system discusses changes in terms of a public application programming interface (API), which specifies what functionality the product offers.

In terms of software, a SQL database’s API could include the types of queries that may be processed. For MCMAS-Dynamic, the public API would include the details of the modelling language it is able to verify properties for. A non-software example could include a simple kettle; the public API could include how one adds or removes liquid, how one turns it on or off, and possibly alarms or other feedback mechanisms for when the liquid has boiled.

When a new version of a product is released, the version number is increased (in lexicographic terms). How this increase is done depends on the types of changes since the previous version:

  • If the public API is ‘broken’, meaning that previously valid ways of using the API are no longer valid or accomplish different things, then the change requires a major version bump. To do this, the major version is incremented, and the minor and patch versions are reset to 0 (e.g. 7.5.1 \leadsto 8.0.0). For example, if the kettle used to play an alarm when it the liquid was boiled and this was a part of the public API, then the major version should be bumped if this functionality is removed. (In particular, if the API did not specify that the kettle had to play an alarm, this change might not warrant a major version bump.)
  • If new features are added without ‘breaking’ the API or there are non-trivial internal improvements, the change leads to a minor version bump. The minor version is incremented and the patch version is reset to 0 (e.g. 7.5.1 \leadsto 7.6.0). For example, if the new version of the kettle is substantially more energy-efficient, then that could be a minor version bump.
  • If something was broken and has been fixed (without changing the public API), then the patch version should be incremented (e.g. 7.5.1 \leadsto 7.5.2). For example, if the kettle previously rang an alarm twice when the liquid was boiled even though the kettle’s API specifies it should only ring once, then a change that makes the alarm only ring once could be part of a patch version bump.
  • Multiple changes in aggregate should be evaluated in aggregate. In most cases, the largest magnitude of all constituent changes applies, though generally speaking this is not true (consider one bugfix plus two changes, one which breaks the API and another that reverts that change – that is a patch bump, not a major bump).

Generally, making a more aggressive version bump than would be required for one’s change is acceptable, though it can confuse users. In particular, I tend to expect backward-incompatible changes when facing a major version bump; not finding any can be surprising and confusing.

The sentiment of the quote sounded like it was a major version bump. Defining an API for one’s life is obviously very difficult; even if one tries to use a lot of abstraction, I find that there are just too many facets. Rather loosely, our API might be split into a bunch of micro-services. We can treat physical needs and bodily functions like breathing and digestion as infrastructural. These services might then focus on the range of activities we involve ourselves in, or range of activities we could involve ourselves in. For me personally, this could include software engineering, getting along with other people, finance and budgeting, computer science, writing, puzzle solving and so on.

Hopefully, I would imagine that we chew through a lot of patch versions as we continue to improve skills. Today’s release notes could include “Jeremy knows a little bit more about thread pools” (I read a chapter of Java Performance Tuning today over a post-lunch coffee). Minor versions would also be relatively common; this wouldn’t be from today specifically, but “Jeremy can vaguely attempt a Balance Loop puzzle” is probably pretty recent, extending the sudoku and other puzzle-solving features.

Depending on how we define the API, major version bumps could be very common. It is typically important to be relatively disciplined with breaking changes in an API in a software engineering context, as clients may often depend on one’s product in non-obvious ways. While others’ dependencies on us can indeed be non-obvious, I think one factor that changes things is that our systems seem to be ephemeral whilst program code is not. A codebase left as it is over years or centuries retains its capabilities (admittedly, finding suitable infrastructure to run the product might be an issue).

On the other hand, there is some evidence that we lose skills that are underutilised with time. I used to play Dance Dance Revolution quite a lot and could probably pass an arbitrary level 15 song as well as some 17s; I doubt I can do that today as I haven’t played in a few years. The ways we interact with others or manage our finances may change as our personal environments change as well; for example, if I moved away from the UK, I would not be able to allocate my investments the way I do now, because I would lose the ability to use ISAs (and probably most other forms of UK-specific tax-free savings). This may even happen without action (for example, if the UK government changes how ISAs or tax-free savings work) – though you could argue that declaring the use of specific vehicles in one’s API might be too specific and implementation-dependent (“I will use tax-advantaged accounts that are valid in my location appropriately” is maybe better).

In light of the above, I would be a bit laxer with what constituted a ‘breaking change’, which pulls things back toward subjectivity which I think semantic versioning was trying to avoid. I might regard myself as having major version 2 right now; I could consider everything up to and including my second year at Imperial as version 0, which is typically used in development to refer to a pre-release period of rapid iteration. Although National Service and/or moving to the UK for studies did bring about nontrivial changes, I really didn’t know what I wanted to do at that time (not that I know now, but there is at least a vague direction).

The Google internship was probably the turning point for version 1; that also coincided with several major changes with regard to finance, investment, philosophy and priorities. I’d call the second major change to be when graduating from Imperial and starting at Palantir; even then, I’d regard the first set of changes to be more fundamental. The re-examination I did over the weekend is actually probably a patch release (or maybe a minor that improves several non-functional characteristics); it certainly doesn’t warrant a major version bump.

Leave a Reply