Take me to where I can buy the sticker!
Unit Testing Is Not A Crime laptop stickers now available on @redbubble.
A lot of people ask me for these but I have always batch-printed them before which meant I couldn’t ship them to random people on the Internet… until now 🌈🦄https://t.co/C71wV1Hhor
I think that if the mental model one is using is like the New View on Systems Safety, then the connections between the disciplines in TOAD (Testing, Observability And Devops 🐸) naturally become obvious as you work more and more with software systems.
Conversely for people rooted in the traditional / Cartesian view of software systems, it’s pretty much impossible to explain. (edited)
If I think that the history of a software is composed of discrete, finite frames of time that can be reconstructed after the fact then good fucking luck explaining to me why I need an observability framework that lets me respond quickly to unknowns.
Because if I think history is composed of discrete, comprehensible frames that can be reconstructed after the fact, then I have a lot to learn from the system out of that history — your observability and prediction techniques are nice but not necessary.
If I don’t accept that history is composed of discrete comprehensible events, then I have very little to learn from history and then the o11y framework and the culture of resiliency makes sense This is btw what is obliquely being gotten at in the Clay Shirkey quote:
Process is an embedded reaction to prior stupidity.
It means people with a lot of process are people with the old / Cartesian view of system safety. They have a lot of process because they think they can predict what sort of events will be important in the future.
Since formal process is reactive, it follows that in order to have a lot of formal process, we have to have a lot of events we can predict with great specificity.
People who operate mainly off of prediction just don’t think that way. We don’t come up with taxonomies of past failures and try to derive ontologies that address them.
Feedback engagement is a metric that describes how often and where developers engage with feedback from the CI system. Engagement rates can be calculated using all of the standard engagement measurement tools from production-facing systems: email open rate, the frequency with which developers respond to slack announcements, and most obviously: the rate at which failing tests are fixed or ignored.
Phenotypic conformance analysis describes the practice of establishing a static analysis ruleset based on the observed properties of the codebase rather than on an existing open source standard or a ruleset voted upon by the team.
Then the initial feedback loop revolves around How much new changesets resemble the code that is already in the codebase. This gets at the important properties that static analysis speaks to: consistency and the absence of syntax errors — without resorting to a normative standard that risks a bad fit with existing practices and can lead to bikeshedding.
Testing is just applied epistemology. — Brett Pettichord
A diagram.
The Kübler-Ross change curve and the Six Phases Of A Project are two time-tested ways of visualizing how organizations cope with change!
Here the Kübler-Ross curve and the Six Phases are together for the first time!
I hope this infographic helps to achieve every initiative in your portfolio!
If testing software and writing code feel very different to you, it’s only because you haven’t written enough code yet. That is only my own admittedly controversial opinion. I believe that everything we call “software testing” is a subset of the activity we call “programming.”
Implementation is a test of a hypothesis. To implement a pattern in code one must first form a narrative or if you prefer a hypothesis. Implementation is itself a test of whether the narrative holds up.
Corollary: Consider that the full specification for a program and the program itself are the same thing. This implies you can’t design computer programs by up-front, complete specification. You are constrained by “the laws of nature” to begin with an incomplete hypothesis and proceed by testing the implementation of said hypothesis, using the results of that test to decide how to go about modifying either the hypothesis or the implementation or both, then repeating that process.
At all levels of the stack, it is always the case that complete
specification is functionally impossible since such a thing could only
exist in the form of a complete implementation. The largest Web
application and the derpiest hello world program both
have this quality: that they cannot be implemented by complete
specification but must be built iteratively via (in)validation of
hypotheses.
For a much more thorough exploration of this idea you can read or refer back to Programming As Theory Building by Peter Naur (1985).
Thanks everyone for upvoting my post on HN!
It’s always exciting to see my work on the front page of Hacker News :)
The software development life cycle is predictable in that any long-lived product will eventually outgrow some of its subsystems. For instance a Web service that begins life with a single monolithic database server will as it scales need the capacity increase that comes from a distributed database. A historical example can be found in Twitter’s original dependence on the ActiveRecord ORM, which over time was replaced with a variety of databases and services.
In one sense this might be considered as a canonical definition of legacy systems: the system contains subsystems that are not optimally suited to day-to-day functioning despite the fact that at some point in the past those same subsystems did function optimally. There is a concept of Software Rot or Bit Rot that metaphorically encapsulates the life cycle phases that precede this sort of legacy system.
The observed course of the software development life cycle is that features begin life in a “working” state — meaning that they satisfy the requirements agreed upon by an empowered group of stakeholders — but that inevitably the same features begin to exhibit bugs that are not in any way related to changes of code nor hosting environment. Software rot (as this phenomenon has come to be called) is caused because the requirements for features continue to change and evolve even after such features are in the hands of their users.
the system contains subsystems that are not optimally suited to day-to-day functioning
This is not a well-understood area of software production, nor does Computer Science have much to contribute by way of solutions. The problems are not algorithmic but environmental, social, aesthetic — in other words it is what programmers like to call a squishy problem because so much of the problem space is taken up not by software but by humans and their co-collaborators.
The idea of engaging with squishy problems is uncomfortable to a lot of programmers. I think this is because programmers currently do not have an opportunity to learn the heuristics that would allow them to distinguish good solutions from bad when it comes to human-and-social issues.
Taking software rot as an example, it is a well-known phenomenon long documented in the literature of software. Yet it has no commonly-agreed up on solution. Its management is not a topic of discussion in job interviews nor in performance evaluations (for the most part). The contraventions for software rot are not listed in general programming books nor taught in coding boot camps.
I do not believe that software rot is ignored as a topic because no one recognizes its importance. I’m pretty sure it’s ignored because no one feels comfortable giving advice about it, because almost no one has successfully dealt with the long-term requirements-changes and subsystem upgrades that go with solving software rot.
The knowledge about how this problem have been successfully dealt with are locked away in a couple of books. And those books are old, using server-side programming and Java as their example environment. That’s a hard sell to a junior engineer fresh out of a boot camp or undergrad program.
The recent movement toward systems thinking in software is a hopeful sign. But we need modern discourse that concerns how to deal with changing requirements over time. And so far such discourse hasn’t been forthcoming despite all the growth and hype about code over the last ten years.
Suboptimization is THE reason for technical debt. But I have to go outside #programming to find discussion of it. https://t.co/cgTN57MrT4 pic.twitter.com/SKjYKUVV4O
— Tentacular Devops 🐙 (@noahsussman)June 18, 2017
The uncomfortable truth is that dev is a complex relationship with an increasingly intelligent other.
The uncomfortable truth is that dev is a complex relationship with an increasingly intelligent other.
— Tentacular Devops 🐙 (@noahsussman)June 28, 2015
—
An illustration of the “funnel” for FOSS developer engagement.
Miller is like awk, sed, cut, join, and sort for name-indexed data such as CSV, TSV, and tabular JSON.
jqjq is currently my tool of choice when it comes to processing all sorts of data. Except XML and CSV. XML is pretty well handled by xmlstarlet but I have never found a CSV parsing tool that I liked. I just make do with using jq to work with CSV and to be honest I find the results I can achieve to be substandard.
Anyway, Miller solves all that and is easy to install even if you don’t use homebrew and need to compile it from source.
mlr --c2j cat my_file.csv | jq .
It is that easy! Now my CSV is structured as JSON, which I have spent a lot of time learning to enjoy working with in jq.