Daniel Compton

The personal blog of Daniel Compton - Projects


I’m thriled to announce that I have finally attained the coveted blue GitHub “Staff” badge!

GitHub staff badge for @danielcompton

I’m joining the Social Coding team at GitHub as a product manager. For the last six years I’ve been writing Clojure professionally; product management is going to be quite different, but I’m looking forward to it. There are a few reasons why I’m looking forward to working at GitHub:

GitHub is the worlds most important social network for developers, and I’m thankful for the opportunity to help build it.

Deciphering IRD’s new acronyms - IIT return and ITN return

This tax season I got an email from IRD reminding me I had to file my return. This happens every year and is usually not a surprise. However this year, the returns were slightly different. They said:

The 31 March 2019 IIT return for {person} is due 8 July 2019.


The 31 March 2019 ITN return for {company} is due 8 July 2019.

I’d never heard of an IIT return or an ITN return. I searched around and couldn’t get any results on Google or IRD’s website search. However, my best guess at the acronyms is:

I think IRD has renamed these in their recent system upgrades and I suspect they’ve renamed all of their returns. If you come across any other acronyms, let me know and I’ll add them here to help others out.

Improving videoconferencing audio quality for remote workers

I’ve been working remotely for about five years full-time. Over that time I’ve talked with colleagues pretty much every work day, so being able to communicate clearly by audio has been crucial. Bad quality audio can quickly turn a good conversation into a frustrating one when you struggle to hear the speaker, or keep needing to ask them to repeat themselves. Videoconferencing is lower bandwidth than in-person communication; I’ve tried to get as good quality when videoconferencing so I can catch as much of what the other person is communicating.

I studied as a musician, and accumulated several pieces of audio gear and knowledge about working with audio which have been very useful for working remotely. A friend asked me for advice about getting some audio gear so I sent him an email. That turned into a post on our company wiki, and now I’m posting it here publicly.

In my experience, for remote videoconferencing to work well, the following things are very beneficial:

Improving audio quality when video conferencing

Here are some tips in rough order of importance for improving audio quality when videoconferencing. You can keep going down the list with diminishing returns of audio quality, stop whenever you and/or your team is happy with the quality you’re getting.

Why Heart of Clojure was special

A few weeks ago I got to attend and speak at Heart of Clojure. I met lots of online friends in person for the first time, and made some new ones too. I’ve thought a lot about how to describe it since then, and every time I come back to the word special.

Others have also posted their thoughts on Heart of Clojure: Fork This Conference, The people’s conference, The hallway track conference, Community with lots of heart, and A courageous conference?.

Here were some things that I think made Heart of Clojure so special. If you’re running a conference, consider stealing some of these ideas.

Heart of Clojure was a very special event. I can’t imagine how much work it was for Arne Brasseur, Martin Klepsch, and all of the other helpers before, during, and after the event but it paid off. Heart of Clojure felt like a very polished event, not something being put on for the first time. Thanks to everyone who organised, helped out, spoke, sponsored, and attended the conference. I hope there is another Heart of Clojure in the future, if you get a chance to go, I highly recommend it.

What do :project/dev and :profiles/dev mean in a Leiningen project?

A few years ago I came across a Leiningen project that defined profiles for :project/dev, :profiles/dev, :project/test, and :profiles/test. It took me a little bit of digging, but eventually I discovered what was happening. This is a convention that I think originated with James Reeves. I’m reposting my issue comment here, so it can be more accessible for searchers.

If you see profiles like this in a project.clj, here is what is happening:

:dev  [:project/dev  :profiles/dev]
:test [:project/test :profiles/test]
:profiles/dev  {}
:profiles/test {}
:project/dev  { ... }
:project/test { ... }

Leiningen has a feature called Composite Profiles. On startup, if Leiningen sees a vector of keywords for a profile, it will lookup each keyword as a profile, and merge them together. For the :dev profile, Leiningen will merge the values of :project/dev and :profiles/dev.

If you want to add any custom settings or dependencies for your own use, you can place them into the :profiles/dev or :profiles/test in your ~/.lein/profiles.clj. If either of these are set in the user’s profiles.clj, they will override the empty :profiles/dev map specified in the project.clj. You need an empty map for :profiles/dev {} in the project.clj, because otherwise Leiningen will complain about a missing profile.

Playing Apple Music from multiple devices on the same Apple ID with a family subscription

When I first started using Apple Music I signed up for a solo subscription. During the day, I would listen to music on my Mac, and my family would listen to music on the iPad. If we listened at the same time, we would get the error:

“Looks like you’re listening to music on another device.”

Eventually this error became too annoying and I looked at upgrading to a Family subscription. The Apple Music Family subscription allows up to six people that are part of the same Family Sharing group to play music at the same time.

It sounded like this was probably what I needed, but in all of the documentation I read, it wasn’t clear whether two devices that were logged into the same Apple ID would be able to play music at the same time. I talked with Apple’s support team and it still was unclear so I went ahead and upgraded to try it for myself.

After upgrading I tested playing on multiple devices, and it worked fine. If you have an Apple Music Family subscription, you’re able to play on multiple devices that are logged into the same Apple ID.

Announcing defn-spec, a library to create specs inline with your defn

I’m pleased to announce the initial release of defn-spec, a library to create specs inline with your defn.

[net.danielcompton/defn-spec-alpha "0.1.0"]

A quick peek at defn-spec:

(ds/defn to-zoned-dt :- ::zoned-date-time
  [instant :- ::instant
   zone-id :- ::zone-id]
  (ZonedDateTime/ofInstant instant zone-id))

One of the features in Schema that I always appreciated was inline function schemas, using a schema.core/defn macro.

When spec was released it had many similarities to Schema, but one thing it didn’t have was a way of expressing specs inline with your function definition. Spec only supported defining function specs separately with fdef. This does have some advantages, it forces you to think carefully about changing your specs, and to be aware of possibly breaking consumers. While this is valuable, not all code is written to these constraints, and I found having fdef’s separate from the function definition had a number of downsides.

When writing Clojure, I noticed that I often resisted writing specs for functions. After thinking about it I realised that I didn’t want to duplicate information from the defn into the fdef. It’s not a huge deal, but it was enough to deter me from writing specs on code that was being heavily modified. This is a really useful time to have basic specs on your functions, so that you can catch refactorings gone wrong early.

I created defn-spec to increase the locality of the spec definitions, and to reduce the activation energy to start adding specs to your codebase. defn-spec copies the syntax (and implementation) of Schema’s defn macro. This has the advantage of adopting a proven design, familiarity for many Clojurists, and the ability to work with existing tooling that understands the Schema defn macro.

Benefits and tradeoffs

Like all things in life, defn-spec has benefits and tradeoffs:



This is similar to Orchestra’s defn-spec macro, but allows for optionally only speccing part of the function, and matches the well-known Schema defn syntax. Try them both out though, and see which one works best for you. defn-spec is still missing some features from the defn macro like destructuring, but they are on the roadmap to be added soon. I’m releasing this early to get feedback from other users.

defn-spec will follow Spec’s release cycle, and there will be a new set of namespaces and artifacts for spec-alpha2 and beyond. If you have features/bug reports, feel free to post them on GitHub.

State of Clojure Survey 2019 Analysis

Cognitect have recently released the results of their State of Clojure Survey for 2019. For the last three Clojure surveys, I have reviewed the free-form answers at the end of the survey and tried to summarise how the community is feeling. This year I’m repeating the exercise, keeping the same categories as before. If you’d like to see all of the comments, I’ve published them in rough groupings.

Some comments have been lightly edited for spelling, clarity, and brevity.

Update: Alex Miller has posted a few responses to some of the comments I highlighted below, and some suggestions for the next steps that people can take to help.

Error messages

Error messages have been a top complaint in Clojure for a long time, I think since the very first survey. CLJ-2373 introduced improvements to error messages in Clojure 1.10. A number of comments complained about error messages this year, but none of the complaints mentioned 1.10’s changes. Given that 1.10 was released on 17 December 2018, and the survey ran from 7-22 January 2019, it seems likely to me that many of the people complaining haven’t tried the latest improvement to Clojure’s error messages.


Spec has been around for two and half years, but is still in alpha. A number of comments referenced Rich’s recent Maybe Not talk about coming changes to Spec. It was great to see Rich putting in to words some of the problems I had felt. It feels like the community is in a transitional phase with spec, where people are still not quite sure where things are going to land.


Documentation has continued to be a sore spot for beginners. API docstrings were often cited as being too hard to understand, instead people preferred to see examples at clojuredocs.org. The documentation at clojure.org has grown this year with a number of great guides added, particularly Programming at the REPL.

Startup time

Startup time has been a perennial complaint, and there haven’t been any major changes here.


Clojure continues to grow in business adoption, but hiring developers has been one of the top three things preventing people from using Clojure for the last three years. I’ll add that one way that companies could address the mismatched supply and demand of Clojure developers is being more open to remote workers. Update: Alex Miller also suggested training people in Clojure, which I thought about when preparing the post, but forgot to include in the final copy.

A new option in the survey this year was “Convincing coworkers/companies/clients”; it was ranked number one in preventing people from adopting Clojure. In my opinion, this is a significant issue to be addressed if Clojure wants to grow its market share further. This probably needs more research to find out what the barriers are to convincing people. Elm is proof that exotic languages can be adopted widely, we should learn from what they’ve done.


There were again more suggestions for improvements to existing language features, or development of new ones.

Language Development Process

2018 had a lot of discussions about Clojure’s development process, and the relationship between the core team and the community. I was very curious to see how widely those feelings were reflected in the free-form comments from respondents. After compliments (~230), this was the most common comment type, with ~70 responses (positive, negative, and in-between) on Clojure’s language development process (out of 563 total comments). While many people had issues with the development process, there were also many people who were supportive of how Clojure is currently developed.


People seem to mostly enjoy the Clojure community, but others have had negative experiences, often citing an elitist attitude from some community members. Losing the history from Clojurians Slack came up several times. I’m not sure if everyone is aware of the Clojurians Slack Log?


People are still looking for a curated on-ramp into Clojure web development, a ‘Rails for Clojure’. There are a number of frameworks and templates here, but they don’t seem to be hitting the spot. I’m not sure whether that is because of limited mindshare/marketing, limited documentation, or the scope and quality of the frameworks. Having used Clojure for many years, I no longer find it difficult to find and assemble the pieces that I need for projects, but I definitely remember finding this being difficult when I started. Data Science and ML were places that people saw a niche for Clojure. Several people hoped for improvements on core.async, it still has a number of rough edges.

Other compilation targets

People have been experimenting with the Graal compiler to produce static binaries without needing a JVM. At this point that seems like the strongest option for people wanting to use Clojure without the JVM. Better Android support was also requested.


For the past few years people have been less and less worried about types. I suspect this is mostly due to spec.


Shadow CLJS was mentioned by many people as having been a great part of their workflow. For those not using Shadow, externs remain challenging.


Clojure’s tooling has continued to improve. This year lots of work has been done adding plugins or expanding functionality of the new official Clojure tools. clj on Windows was asked for by many people. Maintainer burnout seems like a significant risk to the Clojure tooling ecosystem.


As always, the vast number of comments were compliments to Clojure and the core team.


Clojure is a great language, and people are very enthusiastic about it. Its adoption in business continues to grow. There are a number of areas for improvement still, especially if the Clojure community wants to grow further. Not all of this work needs to go through Core though, particularly in areas of documentation and guides, libraries, and tooling.

I worry about the Clojure community losing key contributors. There are a few linchpins holding a lot of things together; if we lose them, it will be hard to come back from. If you don’t want to see this happen then please support the libraries and tools you rely on by contributing code, documentation, issue triage, or money.

On Abstraction

Almost two years ago, there was a Github issue on reagent (a ClojureScript React wrapper), suggesting that Preact be added as a substitute for React. I wrote up a fairly long comment about why I didn’t think this was a great idea (at least not immediately). React’s recent announcement of the new hooks feature made me think about it again. I’ve republished it here with a few edits for context and time.


In principle, I’m not opposed to the idea of Reagent using Preact. It has some cool features and I like that it is small (although in comparison to the total compiled size of a normal CLJS app it’s probably a wash). If Preact worked 100% out of the box with Reagent with no code changes required then I would have no issues with someone swapping out the React dependency for a Preact one and calling it a day. If there are only a few minor tweaks to Reagent required to pass the tests, then again I don’t really have any issues with that. I suspect that even if you have 100% of the tests passing, there will still be issues, as Reagent was built around React, and may not have tests that would cover the difference in behaviour between React and Preact.


It looks like it may not be a pure lift and shift to support Preact. If that’s the case then we run into a bigger issue: abstraction. Reagent was written and built around the ideas and in the context of React. There are assumptions (probably tens or hundreds) built around React’s API and possibly implementation details too.

Adding abstraction adds a large cost because you can no longer program against a concrete API and implementation, you now have to consider two. There are three numbers in computer science, 0, 1, and many. We would be moving from 1 to many, and that takes work.

An aside: recently at work, we were looking at moving a legacy system from only supporting dates in the past to also be able to support dates in the future. This should be straightforward right? We talked to the programmers responsible for it and they couldn’t guarantee that it would work, nor whether supporting future dates would be easy or hard. In the building of that (or any) system, hundreds of simplifying assumptions are made around the context that the system is going to be built in.

It is a very common pattern to have different backends and I don’t see any downsides to it.

I can’t think of a single example of a system with multiple backends that didn’t have any downsides to it, e.g. ORMs, HTML/CSS/JS, Java. There may be some, but they would be the exceptions that prove the rule. Everything has a cost, the question is whether there is a benefit that outweighs the cost. It is much harder to remove something from software than to add it, which is why we should be certain that the benefits outweigh the costs.

While Preact strives to be API-compatible with React, portions of the interface are intentionally not included. The most noteworthy of these is createClass() … https://preactjs.com/guide/switching-to-preact#3-update-any-legacy-code

Reagent currently uses createClass. There are workaround options provided, but this is an example of some of the API differences between React and Preact which you need extra compatibility layers to support. Do we know if the compatibility layer works 100% correctly?

A possible future if Preact support is merged now

As a thought experiment, let’s assume that Preact is in Reagent with some kind of compatibility shim. Preact already has several performance optimisations that people can take advantage of:

customizable update batching, optional async rendering, DOM recycling and optimized event handling via Linked State. - (from Preact homepage)

Wouldn’t you want to be able to take advantage of those in your application? I certainly would. Now to do so, you may run into issues because the compatibility shim layer that was written was encoded around default assumptions of React, and they may not apply to Preact. Do we have to rework the shim layer, or lower level Reagent API stuff? Who is going to do that work? Who is going to review it and merge it?

Let’s consider the reverse. Perhaps in some new React version, Facebook comes out with a new API which is faster or better suited to Reagent’s style of rendering, so we want to switch to that [since writing this they came out with Hooks. These may be added to Preact also, but this isn’t certain and neither is the time-frame]. However that new model may not work with Preact. Again, we’re in a bit of a pickle: Preact users want to be carried along with Reagent and get the benefits of new Reagent work, but it may not be easy or possible to support the new API for them. Now what?

Consider everyday development on Reagent. Reagent’s source code is built around a very detailed understanding of React and is highly optimised. If Preact was supported too, then developers would probably need to gain an understanding of Preact too.

At the moment, Preact has one main contributor, it has been around for 1.5 years. React has many contributors. I’d estimate there are 100+ people with very deep knowledge of React. It’s been around (in public form) for 3.5 years. In general, the JavaScript community does not have a reputation for long-term support of projects. What happens if development slows/stops on Preact and the compatibility layer isn’t kept up to date? It is much harder to remove something than it is to add it. Who decides when/if to remove Preact from Reagent at a future date?

These are all hypotheticals, but I hope this demonstrates that the extra abstraction provided by supporting two VDOM layers doesn’t come for free. At the very least, it consumes extra brain cycles when testing and developing Reagent, extra support and documentation costs from users wanting to use one or the other, as well as extra indirection when running and debugging apps using Reagent.

The Innovators Dilemma

If you haven’t already, I highly recommend reading “The Innovators Dilemma” by Clayton Christensen. One of the key points he makes in that book is the difference between integrated and modular products and when to develop each kind.

CHRISTENSEN: When the functionality of a product or service overshoots what customers can use, it changes the way companies have to compete. When the product isn’t yet good enough, the way you compete is by making better products. In order to make better products, the architecture of the product has to be interdependent and proprietary in character.

In the early years of the mainframe computer, for example, you could not have existed as an independent contract manufacturer of mainframe computers, because the way they were made depended upon the art that was employed in the design. The way you designed them depended upon the art that you would employ in manufacturing. There were no rules of design for manufacturing.

Similarly, you could not have existed as an independent maker of logic circuitry or operating systems or core memory because the design of those subsystems was interdependent. The reason for the interdependence was that the product wasn’t good enough. In every product generation, the engineers were compelled by competition to fit the pieces of the system together in a more efficient way to wring the maximum performance possible out of the technology that was available at the time. This meant that you had to do everything in order to do anything. When the way you compete is to make better products, there is a big competitive advantage to being integrated. … In order to compete in that way, to be fast and flexible and responsive, the architecture of the product has to evolve toward modularity. Then, because the functionality is more than good enough, you can afford to have standard interfaces; you can trade off performance to get the advantages of speed and flexibility. These standard interfaces then enable independent providers of pieces of the system to thrive, and the industry comes to be dominated by a population of specialized firms rather than integrated companies.

I would argue that we are still very much at the point where the current VDOM libraries aren’t good enough yet. They aren’t yet ready to be commoditised, and the best option is to tightly integrate.

Options from here (with some conjecture)

  1. Someone can make a PR to Reagent to add support for Preact. It will probably take a while to get merged because it is a significant change. Once it is merged and released, there will probably need to be several rounds of revisions before it is ready to go. Because Reagent moves relatively slowly, this will take a while.

    Reagent also has a large number of production users, so new releases need to be well tested and stable. Adding Preact to the mix is going to slow this down further.

  2. Someone can make a fork of Reagent (let’s say it’s called Preagent). You can run wild experimenting with what is the best way to use Preact in Preagent, take advantage of all of the great features Preact has, and have a much faster turnaround time for releasing and using it. You will be able to work out what is the right API and integration points for Preact because you have room to experiment with it, without the weight and responsibility of bringing the rest of the Reagent users along with you.

    At some point in the future, you could review merging Preagent back into Reagent, given all that you now know. You would also have the weight of evidence on your side where you can demonstrate the benefits of Preact and can show how many users want Preact. This would let you make a much better case for including Preact, give you what you want in the meantime, and likely provide a higher quality integration in the future.

    Alternatively, you may decide that Preagent is better served going its own way and integrating more closely with Preact. This is also a good option.

Abstraction is not free

The point I have been trying to drive through this post is that abstraction is not free. Over-abstraction is a common anti-pattern and it saps productivity. I had a friend who recently left a Clojure job and started a Java one. He quipped to me about how he’d forgotten what it was like to trace code through five layers of abstraction to get to the concrete implementation. As programmers, we’re trained to solve problems by adding layers of abstraction, but that isn’t always the best way to solve the problem.

Lastly, this isn’t a personal attack on you or your ideas. I’m all for innovation in ClojureScript webapps and I think that it is worth investigating Preact and how it could work in the ClojureScript ecosystem 😄. I’m not against Preact. I would consider using it if there was a measurable benefit. I’m just suggesting that the best way to go about this is probably not to integrate it into Reagent as the first step.

Announcing The REPL podcast

I sent the first newsletter for The REPL two years ago, and I have really enjoyed writing it. I’ve learnt a lot from what people have written and created to share with the Clojure community. Sometimes when I’m writing the newsletter I’ve thought “That’s fascinating, I’d love to hear more about the technical details of that”. Now I have an outlet for doing just that.

I first had the idea of producing a Clojure podcast around January 2018. I didn’t have the time for it then, but the idea kept swirling around the back of my brain. When Michael Drogalis’s Pyrostore was acquired, I wanted to hear more about the acquisition, so I contacted Michael to see if he wanted to talk about it. That became the first episode.

My goal for The REPL is to talk with people about the technical details of the projects they are working on. The Clojure community has a ton of interesting people doing really creative work with Clojure. I’ve really enjoyed talking with a bunch of people already, and I’m looking forward to talking to more of them in the future. You can find it therepl.net. It’s available on Apple Podcasts, RSS and all of the other common podcasting apps.

State of Clojure Survey 2018 Analysis

Cognitect has recently released the results of their State of Clojure Survey for 2018. For the last two Clojure survey’s, I have reviewed the free-form answers at the end of the survey and tried to summarise the zeitgeist of community feeling. I enjoy it each time, and so here’s this years analysis.

Some comments have been lightly edited for spelling, clarity, and brevity.

Error messages

Error messages have been one of the top complaints about Clojure since the surveys started, and this year they have gotten worse with the introduction of specs on core macros. I’m not the only one who has noticed this either. In 2015 Colin Fleming gave a talk on improving Clojure’s error messages with Grammars. In previous surveys there was hope that spec would be able to use this approach to improve error messages; Ben Brinckerhoff has recently shown a proof-of-concept that gives some similar (excellent) error messages in Expound.


Comments ranged from loving spec to disliking it. There were also requests for more guidance on how to apply spec to real-world use-cases and how to integrate it into whole programs.


Documentation in all forms has been another perrenial complaint but this is starting to improve. clojure.org continues to add more guides this year on Programming at the REPL and Higher Order Functions among other things. Martin Klepsch’s cljdoc holds a lot of promise to improve the state of third-party documentation, by automatically generating and publishing docs for published JARs.

Startup time

Startup time continues to be an issue for people. As serverless computing becomes more mainstream, this may end up pushing people from Clojure to ClojureScript, or elsewhere.


There were lots of comments this year in this category. Even after 10 years, Clojure still unknown or seen as being extremely niche to many people, especially among business stakeholders. Hiring seems to be less of an issue than in years past. There is still a large contingent of programmers looking to work with Clojure in remote positions, although it’s hard to say whether this proportion is any higher than in other language communities.


Language Development Process

This hasn’t changed in a long time, and I don’t see it changing in the future, but it has been persistently highlighted by many people as a big issue.


By and large, people found the community welcoming, and personally I have found it to be one of the best things about working with Clojure. However there is a persistent undercurrent of eliteness from some Clojure programmers which really puts people off the language.


More guidance on how to compose and select libraries remains a common common issue for people in this section, as well as improving the state of library documentation.

Other targets

As in previous years, there are a smattering of people asking for alternative platforms to target other than JS and the JVM. LLVM, Go, and Web Assembly all had interest from people.


Requests for static typing were down from previous years. I attribute that mostly to spec gaining wider adoption and understanding.


ClojureScript continues to improve, although getting the tooling setup remains a pain point, as does integrating NPM modules. There are recent improvements on that front though, so this may not be such an issue in 12 months time. I have heard lots of good things about shadow-cljs, it seems to be a strong tool to investigate for ClojureScript usage, especially if you work with a lot of NPM modules.


Tooling on the Clojure side seems to be somewhat less of an issue this year, but setting up ClojureScript projects with all the things that you need remains an issue for many people.


As always, the survey was dominated by positive comments. Here were some of my favourites.


When I went to write this years summary, I re-read the previous posts to remind myself of where we were a few years ago. While some areas have improved, I was struck by how many of the comments could have been lifted from previous years survey results. There was a worrying trend in the community and adoption sections that people perceive Clojure to be shrinking. Perception doesn’t always match reality, and it may be (as a commenter noted) that Clojure is entering the trough of disillusionment phase of the Gartner Hype Cycle. However I think there are a lot of low-hanging fruit around that could significantly improve the experience and ecosystem around Clojure without changes to the core language.

How to serve ClojureScript files in development

When I develop ClojureScript projects, I almost always use Figwheel. It’s a great tool, but sometimes my app ended up using stale files. This led to some very confusing debugging sessions. It only happened some of the time, and was always fixed after a hard refresh. I thought about just disabling the browser cache, but I didn’t like ignoring the issue. After seeing colleagues struggle with stale caching too, I decided to figure out what was going on, and fix it once and for all.

Cache-Control rules everything around me

The first thing to do was to add a Cache-Control: no-cache header to all static file responses. Despite the name, no-cache tells the browser it can cache files, but must always validate them with the server before using them. If the browser’s cached version is up-to-date, a compliant HTTP server should return a 304 Not Modified response, otherwise it serves the new file.

If you don’t provide a caching header to an HTTP response, the browser can choose its own caching behaviour. The browser’s caching heuristics are much more aggressive than you want in development, and lead to the weird caching behaviour I was seeing.

I thought this had fixed the issue, but occasionally I would still notice stale files were being used. After looking closely at the compiled output files, I made a surprising discovery.

ClojureScript copies file modification times

ClojureScript (as of March 2018) copies the last-modified date of ClojureScript source files to the compiled JavaScript target files. This is so that the compiler can detect changes to source files. JavaScript from the Closure compiler (e.g. goog.base), gets a modification time that matches the time it was compiled.

Neither of these dates are particularly useful to use as a Last-Modified date header for caching purposes.

To avoid these issues, I recommend removing the Last-Modified header from the response when in development.


To knock both problems on the head once and for all (hopefully), I added a CRC32 checksum based ETag for static file responses. I packaged this up in a library ring-etag-middleware so that other projects could also use it.

Serve 304 Not Modified responses

At this point the browser will check with the server for every ClojureScript file, on every pageload. However, this causes all of the files to be downloaded each time, even if they haven’t changed. The last step is to add ring’s ring.middleware.not-modified/wrap-not-modified middleware. This returns a “304 Not Modified” response if the ETag provided in the If-None-Match request header matches the ETag header in the response.


As best as I can tell, this has completely solved all of the odd caching issues that I was seeing, while still keeping the app snappy to load by reusing as much of the cache as possible. If you are serving ClojureScript files in development and not using Figwheel, I recommend you follow these three steps:

  1. Set a Cache-Control: no-cache header
  2. Add an ETag to your static file responses
  3. Remove the Last-Modified header
  4. Wrap your responses in ring.middleware.not-modified/wrap-not-modified or the equivalent in your Clojure web framework.

Adding Context to CockroachDB’s Article “Your Database Should Work Like a CDN”

I was excited this morning when I checked Hacker News and saw an article from Cockroach Labs titled “Your Database Should Work Like a CDN”. I’m a big fan of Cockroach DB and enjoy their blog posts. I clicked on this one, but came away disapointed. I don’t think it fairly contrasted Cockroach with it’s competitors, and instead made some incomplete arguments instead of selling Cockroach on it’s merits.

In this article I will analyse sections of the post and add more context where I think some was left out.



To maximize the value of their services, companies and their CTOs chase down the elusive Five Nines of uptime

Only companies with sophisticated operations teams can seriously set an SLA for five nines. It’s doable, but comes with a heavy lift if your service is doing anything non-trivial. It’s certainly not the default position among the industry as far as I can tell. The Google SRE book has a good section on this.

For many companies the cost of moving from a single region deployment to a multi-region one is too great and doesn’t provide enough benefits. Some services don’t actually need to be available to five nines, and if not done well, moving to multi-region deployments may make your system less fault-tolerant, not more.

This is particularly crucial for your customer’s data. If your data’s only located in a single region and it goes down, you are faced with a “non-zero RPO”, meaning you will simply lose all transactions committed after your last backup.

If a whole region dropped into a sinkhole in the earth then you would lose all transactions after your last backup. However that’s not a failure scenario that we tend to worry about (although maybe we should?). Every time a major cloud provider has had a zone/region go down, data was partially/wholly unavailable during the outage, but no committed data was lost once service was restored (that I know of?).

…and Bureaucracy?

This impending legislation requires that businesses receive explicit consent from EU users before storing or even processing their data outside the EU. If the user declines? Their data must always (and only) reside within the EU. If you’re caught not complying to GDPR? You’ll face fines of either 4% of annual global turnover or €20 Million, whichever is greater.

I’m not a GPDR expert, but this seems like an incomplete summary of the GPDR rules around processing, especially as the processing rule applies only to the processing of an EU user’s personal data (AFAICT), and conflates requiring consent for storing/processing personal data with storing the data outside of the EU. From article 6:

Processing shall be lawful only if and to the extent that at least one of the following applies:

a) the data subject has given consent to the processing of his or her personal data for one or more specific purposes;

b) processing is necessary for the performance of a contract to which the data subject is party or in order to take steps at the request of the data subject prior to entering into a contract;

c-f) [Public/government interest exceptions)

If you have entered into a contract with the data subject, and that data is necessary for the contract, then I think you are fine (IANAL). As far as I can tell, transfers to Third Countries are OK as long as there are appropriate safeguards.

This is the part of my critique that I am least confident about, I’d welcome some links confirming/rebutting this.

When you take GDPR in the context of the existing Chinese and Russian data privacy laws––which require you to keep their citizen’s data housed within their countries […]

If you follow the links in the article, you see this for China:

Does the Cybersecurity Law require my company to keep certain data in China?

[…] To that end, the Cybersecurity Law requires “critical information infrastructure” providers to store “personal information” and “important data” within China unless their business requires them to store data overseas and they have passed a security assessment. At this point, it remains unclear what qualifies as “important data,” although its inclusion in the text of the law alongside “personal data” means that it likely refers to non-personal data. […]

“Critical Information Infrastructure” providers are defined a bit more narrowly, but the law still casts a fairly wide net. […] the law names information services, transportation, water resources, and public services, among other service providers, as examples.

and this for Russia:

3.6 Generally, to transfer personal data outside the Russian Federation, the operator will have to make sure, prior to such transfer, that the rights of personal data subjects will enjoy adequate and sufficient protection in the country of destination.

Some companies will need to store data on China/Russian servers, but the laws here are far narrower than “If you have any data from a Chinese or Russian person or company, you must store it in their country”.

Managed & Cloud Databases

Managed and Cloud databases often tout their survivability because they run in “multiple zones.” This often leads users to believe that a cloud database that runs in multiple availability zones can also be distributed across the globe.

It might be an assumption you would make if you have no experience with databases, but it’s not one that I’ve ever seen a software engineer make.

There are caveats to this, of course. For example, with Amazon RDS, you can create read-only replicas that cross regions, but this risks introducing anomalies because of asynchronous replication: and anomalies can equal millions of dollars in lost revenue or fines if you’re audited.

I’m not sure how asynchronous replication lag could result in failing an audit and incurring millions of dollars of fines. I spent a few minutes trying to come up with a scenario and couldn’t. Losing revenue from users also seems speculative, I’m not really clear how this would happen.

Designing a system to run across multiple regions with asynchronous replication is certainly not trivial, but people do it every day. If they were losing millions of dollars from it, they would probably stop.

In addition, this forces all writes to travel to the primary copy of your data. This means, for example, you have to choose between not complying with GDPR or placing your primary replica in the EU, providing poor experiences for non-EU users.

Again, I don’t think GPDR requires this.


For example, NoSQL databases suffer from split-brain during partitions (i.e. availability events), with data that is impossible to reconcile. When partitions heal, you might have to make ugly decisions: which version of your customer’s data to you choose to discard? If two partitions received updates, it’s a lose-lose situation.

This paints all NoSQL databases with a broad brush. While some NoSQL databases are (were?) notorious for losing data, that is not inherent to NoSQL databases.

Certainly if you are using an AP NoSQL database, you need to design your application to correctly handle conflicts, use CRDT’s, or make idempotent writes. Partitions do happen, and it’s not trivial to handle them correctly, but neither is it the case that you always need to discard your customer’s data.

Sharded Relational Databases

Sharded Relational databases come in many shapes and suffer from as many different types of ailments when deployed across regions: some sacrifice replication and availability for consistency, some do the opposite.

I assume the author is referring to systems like Citus. I don’t have enough experience with systems like this to judge the assertion, but this seems fair.


If you do need more reliability/availability than is possible from a single region, then Cockroach is a strong option to consider. I think a far better argument for CockroachDB and against NoSQL, Replicated SQL, and Sharded Relational DBs is minimising complexity and developer time. It is possible for developers to design their application for the nuances of each of these databases, but it’s certainly not easy or cheap, especially if you want it to be correct under failure. The reason Google created Spanner (the inspiration for Cockroach) was that developers found it hard to build reliable systems with weak consistency models.

[…] It is much easier to reason about and write software for a database that supports strong consistency than for a database that only supports row-level consistency, entity-level consistency, or has no consistency guarantees at all. - Life of Cloud Spanner Reads & Writes

CockroachDB provides consistent reads and writes, supports SQL, is able to be deployed multi-region, and in any datacenter. Those are a compelling set of features. If your application can handle the latency and performance tradeoffs that it makes (which are getting better all the time), then it will let you write software against a consistent datastore without spending as much time reasoning about hard consistency problems. Cockroach is a great product, and I think it stands well on it’s merits.

How to upgrade Terraform provider plugins and modules

Since Terraform v0.10, Terraform providers are distributed separately from the Terraform binary. This lets them update at different paces, and allows a wider group of people to collaborate on the providers. This is mostly good, but it does introduce a new step for upgrading providers. It is slightly counterintuitive, but to upgrade your providers, run

terraform init -upgrade

To upgrade your modules, run

terraform get -update

Fixing Ansible error: “SSH Error: data could not be sent to remote host "". Make sure this host can be reached over ssh”

I was building VM images for Google Cloud with Packer, and provisioning them with Ansible. Everything had been working in the morning, but in the afternoon one computer wasn’t working after I had upgraded Ansible with Homebrew. I was having a really tough time figuring out why Ansible and Packer were running fine on one computer, and not on the other. I was getting the following error:

googlecompute: fatal: [default]: UNREACHABLE! => {"changed": false, "msg": "SSH Error: data could not be sent to remote host \"\". Make sure this host can be reached over ssh", "unreachable": true}

Investigating the error, I found it could indicate any number of problems, as Ansible is masking the underlying SSH error.

After checking the versions of Python, Ansible, and Packer on the two computers, I found one difference. On the computer that wasn’t working, when running ansible --version it had a config file listed:

$ ansible --version
  config file = /Users/<user>/.ansible.cfg
  configured module search path = Default w/o overrides
  python version = 2.7.13 (default, Dec 18 2016, 07:03:39) [GCC     4.2.1 Compatible Apple LLVM 8.0.0 (clang-800.0.42.1)]

Opening that file:

$ cat ~/.ansible.cfg
pipelining = True

I moved that file to another location so it wouldn’t be picked up by Ansible, and after rerunning Packer a second time, got this error:

googlecompute: fatal: [default]: UNREACHABLE! => {"changed": false, "msg": "Failed to connect to the host via ssh: Warning: Permanently added '[]:62684' (RSA) to the list of known hosts.\r\nReceived disconnect from port 62684:2: too many authentication failures\r\nAuthentication failed.\r\n", "unreachable": true}

I then used the extra_arguments option for the Ansible provisioner to pass [ "-vvvv" ] to Ansible. I ran this on both computers and diffed the output. I saw that the dynamically generated key was being successfully provided on the working computer (after one other local key). On the failing computer I had many SSH keys that were being tried before I could get to the dynamic key, and I was getting locked out.

SSH servers only allow you to attempt to authenticate a certain number of times (six by default). All of your loaded keys will be tried before the dynamically generated key provided to Ansible. If you have too many SSH keys loaded in your ssh-agent, the Ansible provisioner may fail authentication.

Running ssh-add -D unloaded all of the keys from my ssh-agent, and meant that the dynamic key Packer was generating was provided first.

I hope this is helpful to someone else, and saves you from hours of debugging!


I was very confused by seeing that my computer was trying to connect to, instead of the Google Cloud Platform VM. My best guess is that Packer/Google Cloud SDK proxies the connection from my computer to the VM.

Detecting the users’s time zone using pure JavaScript

While working on Deps I wanted to detect the user’s time zone at signup, to localise the times that I presented to them. I hunted around and found a variety of libraries that offer time zone detection, like moment.js, and jstz. However I wasn’t keen on pulling in a dependency on one of these libraries if I could help it, so I kept looking. I also considered trying to detect the user’s time zone from their IP address using something like the Maxmind GeoIP database, and that probably would have been the method I settled on.

However in my research, just before I started with IP based detection, I found that there is a newish Internationalization API, which will give you the user’s system time zone.

=> "Pacific/Auckland"

All current browsers support the Internationalization API including IE11. However there is a slight wrinkle to keep in mind when using it: on some older browser versions that do support Internationalization (IE <=11, and Firefox <=51), the timeZone property is set to undefined rather than the user’s actual time zone. As best as I can tell, at the time of writing (July 2017) all current browsers except for IE11 will return the user’s actual timezone. If you do need to support older browser environments then you could look at moment.js. It uses the newer Internationalization API when it is available, and falls back to other heuristics if it isn’t. For my purposes, my browser statistics show almost no IE usage, so I chose to just use the newer method, while also allowing the user to manually update their time zone if they needed.

Bryan Cantrill on Integrity

Bryan Cantrill on integrity. Edited slightly for clarity and profanity:

We are on a collision course with the Amazon Principles […] What is going on? I’m losing my mind. The wheels are off, everyone has to get out of the car and this is why. All of these [expletive] leadership principles, from all these organisations, where is integrity? Damn it where’s integrity? Amazon has 14 leadership principles and integrity is not on it. [That’s] inexcusable. I’m sorry, if you’ve got one principle in your organisation, its integrity? Right? […] No we’re living in a world that has lost it’s [expletive] mind, I don’t understand it. Why the appetite for territory? Do you not know where your next meal is coming from? Do you not have a roof over your head? I mean I’m sorry, I just don’t get it. We have got an incredible luxury. There’s never been a labour market like this one. Where have we so screwed up with a generation, if not a society where we’ve got people who are so extrinsically motivated? What the [expletive] is wrong with us? I mean its like how is integrity not the only thing you have?

[Andrew Clay Shafer]: Are we doing the full critique of capitalism today?

No it’s not. Bullshit, it’s not capitalism. The finest capitalist is Scott McNealy, a capitalist so pure that when he was being devoured by Oracle he believed that it was his duty to capitalism to die, and I admire that. There is no purer a capitalist than Scott McNealy. And read McNealy’s final email to Sun employees, and it’s almost prescient, McNealy says: “You know what, in 30 years I never had to hide the newspaper from my children.” An achievement that Uber violated on like month two? Where are Travis’ parents? Are you not humiliated? How did you raise him to be so divorced from what really, actually, truly matters? What the living [expletive] is wrong with us? Maybe it’s me?

I continue to be impressed by Joyent, and Bryan Cantrill’s engineering principles. It sounds like a great place to do your best work.

Recorded at GOTO 2017 for an Arrested Devops episode: Old Geeks Yell At Cloud

The case for and against Clojure predicates returning nil

Last week, there was a mailing list post on the clojure-dev mailing list noting that several of the new predicates in Clojure 1.9 alphas return true or nil. The three new predicates are: qualified-keyword?, qualified-ident?, and qualified-symbol?. Previously, all predicate functions in Clojure core, i.e. functions that end in a ? have returned true/false. In this post I want to examine the pros and cons of this new behaviour.

If I’ve missed anything here, please get in contact and I’ll update the post.

Pros to some predicates returning nil

The first two points aren’t strong arguments for returning nil, rather they argue that it doesn’t matter whether the functions return nil or false.

Cons to some predicates returning nil:


Personally, I don’t see a strong reason to keep the current behaviour of the new qualified predicate functions, but you might see things differently. Whatever the outcome, I’m glad that in Clojure we don’t have to name all of our boolean functions with an “Eh” suffix.


Alex Miller suggested to file a ticket about this behaviour, so I opened CLJ-2141.

Uber-Alphabet Lawsuit Updates

Since my post a few days ago on the Uber-Alphabet lawsuit, there have been a few new developments.

Levandowski consulting for Uber

In that post I wrote:

… no funding appears to have been taken from Uber, or any other company (If there turn out to be payments from Uber to Otto prior to the acquisition then that would be Very Bad for Uber).

Not long after, Johana Bhuiyan at Recode posted a story which suggests that Uber CEO Travis Kalanick hired Otto CEO Anthony Levandowski as a consultant for Uber’s self-driving car around February-March 2016. Even after reading the article several times and sleeping on it, I still can’t quite believe it.

Now, sources are adding more to that claim. Kalanick hired Levandowski as a consultant to Uber’s fledgling self-driving effort as early as February 2016, six months before Levandowski publicly got his company off the ground, according to three sources.

If this was a legitimate business arrangement, the timing makes no sense at all. Levandowski had just left Waymo weeks earlier. While it might have been difficult for Alphabet to bring a lawsuit on these grounds, the optics for Levandowski advising Uber are terrible. Additionally, Uber and Otto are competitors in the self-driving vehicle space, and Uber is famously aggressive. That a CEO of one company would be advising one of their direct competitors within months of forming their own self-funded startup is truly strange.

If this wasn’t a legitimate business arrangement, then it also doesn’t make much sense purely because it looks so suspicious, especially in light of Alphabet’s trade secrets lawsuit. It brings the start of Uber’s direct involvement with Otto to within weeks of Otto being formed. If the plan all along was for Otto to join Uber, then it does make some kind of twisted sense for him to be advising the Uber self-driving car team.

A little birdie tells me that we haven’t seen the last on this particular thread of the story.

Naked Capitalism on Uber

Hubert Horan has just published another article in his series on Uber and whether it can ever increase overall economic welfare.

The answer is that Uber is attempting to implement an innovative and disruptive strategy for achieving market control — a strategy radical different from every prior tech oriented startup — but those disruptive innovations have nothing to do with technology or the efficient production of urban car services.

Otto purchased another Lidar company

A reader sent me a link to this article by Reuters discussing an acquisition of Tyto LIDAR. The title is “Uber’s self-driving unit quietly bought firm with tech at heart of Alphabet lawsuit”. The title isn’t accurate, as it was Otto that acquired Tyto (for an undisclosed sum) in May 2016. Uber then acquired Otto/Tyto in August 2016.

Levandowski: “We did not steal any Google IP”

Forbes published an article on 28 February, 2017 which contains part of an interview they did with Levandowski in October 2016.

Q: The question is, Google has been working on this for however many years. You guys (Otto) are a year old or something like that.

A: We started in January. How did we catch up?

We did not steal any Google IP. Just want to make sure, super clear on that. We built everything from scratch and we have all of the logs to make that—just to be super clear.

But this is my fourth autonomy stack. I did the DARPA Grand Challenge.

I wonder if “we have all of the logs” is also going to be Alphabet counsel’s opening argument?