The Hipster Stack!

This post describes my version of the Elm hipster stack. The original setup I found is located on github at carleryd/elm-hipster-stack. I had trouble with that version out of the box, because it wasn’t originally generated with Phoenix version 1.3, but rather was converted.

Besides, I think it is instructive to see how this stuff is put together.

Currently my version of the hipster stack does not yet include GraphQL. I’ll get to it when I get there.

  • Elm(version 0.18)
  • Phoenix(version 1.3)
  • GraphQL (TBD)
  • PostgreSQL (9.6.2)

How to rather than git clone

Because it is dumb to just copy and paste this, and because any “official” hipster stack will quickly get out of date given that the components (Phoenix and Elm) are rapidly evolving, what follows is how to get phoenix and elm up and running in the same repo. GraphQL might follow later, but I’m not using it that much yet (until I get the hang of phoenix and elm, it seems a bit too much to learn all at once).

Create a phoenix app thingee

Do something like

export MYAPP="hello_hipster_stack"
mix $MYAPP

Set up the database

Continue reading


Learning about blockchain; starting with some ICO events

I’ve decided to dive deeper into blockchain and Etherium. Because I’m a transportation guy, I did a search for ICOs in the transportation space and came up with TrafficX, ZeroTraffic, and DOVU. I’ve ranked them from sucktastic to slightly silly.


TrafficX looks like a complete scam. I can’t see anything solid anywhere, and their development plan seems to say “Pay us a lot of Ether and we’ll hire some developers. Get in early!” Their initial offering has closed and they raised some money. If it isn’t a scam, we should see some results soon. If it is a scam, I guess the value of the ICO will eventually drop to zero.


ZeroTraffic at least has people behind it who have been in the transportation space for years, but I think their idea is weak. From my read-it-once-through, not super-careful reading of their whitepaper, they are basically using their ICO to allow people to buy “deputy mayor of traffic” positions for their local areas. They will use the funds from their ICO to pay for further development of a website and apps for the gamification of traffic choices, and will use the (hopefully relevant and insightful) input of the local traffic mayors to get people to travel in off-peak times. In my opinion, this is a dumb idea, will not at all result in “Zero Traffic”, and is a bad investment on the part of the ICO investors. On the other hand, the team is honest about their idea, and will make a good faith effort to move the idea from concept to reality. This ICO certainly isn’t a scam.


Finally, DOVU has big rhetoric but more prosaic (and achievable) goals. Their slogan is “Blockchain powered mobility”, but in their whitepaper it turns out they are just trying to monetize the sharing of traffic information. I get the feeling they were casting about for a good idea in the mobility space, and came up with this after several brainstorming sessions. But I also get the feeling that they have mostly computer scientists on their team, rather than transportation economists. They are correct that their technology will likely make it possible to monetize sharing data about one’s trips, but I think they are incorrect in their presumption that this information will have any intrinsic value.

This blind spot arises because we as a society are still transitioning from a world of scarce traffic information to one of plenty. Just ten years ago it would have been unthinkable for a transportation engineer to expect to have accurate, up to the minute estimates of travel times on city streets. Major highways have been well instrumented for decades using loop detectors, but even that data can be spotty and is notoriously bad at estimating speeds and travel times. In the early days of intelligent transportation systems (ITS) there were many business ideas floated based on the idea that specially instrumented probe vehicles with wireless data transponders could be used to collect traffic data. That travel data could then be sold to travelers who wished to know the latest and greatest traffic conditions and the best routes to their destination. Today it is possible to view current traffic conditions for both highways and the major surface streets, and services like Google maps will (for free) provide turn by turn navigation that includes advice about traffic incidents, the expected severity of traffic incidents, and whether a better alternative route exists.

(As an aside, I never believed any of that was necessary, and I wrote a paper back in 2002 in which I argued that regular people could just post traffic conditions on roads much like bloggers post reviews of restaurants. It was rejected even for presentation at TRB that year with a scathing review. I was pissed because the review was so obviously biased towards the current status quo (big top-down ITS projects), and so after that I just gave up on writing papers and publishing in transportation. Ironically, I also kinda described how Waze works with that paper, which would make me feel vindicated if I had gone on to found Waze and now had millions of dollars in the bank, but I didn’t and I don’t, so I’m just bitter instead!)

Anyway, back to the critique of DOVU, now that every vehicle with a cellphone is a traffic probe data collector, DOVU looks at that as an opportunity to “get paid to drive” by selling the information you collect. Unfortunately, I think they are unintentionally mixing up both the past and the present. In the past, traveler information was scarce and therefore valuable. Today, when every cellphone is broadcasting location data to the wireless provider, to the phone’s OS itself, and to any app that is running with permission to track location, traveler information is abundant, not scarce, and therefore is worth very little.

The downsides for DOVU are two-fold. First, even if DOVU operated in a vacuum, if everybody were to participate in their system, the basic rules of competitive markets dictate that marginal cost pricing will hold, and the marginal cost of one additional traveler’s information is next to zero. Second, DOVU does not and will not operate in a vacuum. It is trying to break into a space in which many players are already collecting a wealth of traveler information, with no way to force the current market players to use their system. There is nothing in the DOVU setup that will in any way prevent the many actors who are already gathering traveler information to continue to do so. All that information that is currently circulating without DOVU will continue to do so outside of the DOVU walled garden, keeping the same arrangement that providing that information is part of the cost of using a phone, getting turn by turn directions, and so on. DOVU participants will be competing to sell their traffic information with each other as well as with everybody outside of the DOVU ecosystem (who are basically giving their information away for free). Thus there will be next to zero benefit to any purchaser for any DOVU user’s travel data, and so the market clearing price of all this collected and monetized information will be zero.

To their credit, DOVU seems to be thinking about information beyond just travel times and speeds. In presentations and in the white paper, they imply accessing the vehicles’ information systems to gather data such as using windshield wipers, emergency braking activation and so on, and suggest that those secondary measurements can be used to reveal hidden truths such as weather conditions and traffic hazards. In practice, I don’t think so. We have weather stations for the former, Waze users for the latter. There might be some facet of travel that could be measured that might be interesting, but I doubt very much that anything would be valuable enough to satisfy DOVU’s implied promise that drivers could get paid to drive.

All those criticisms aside, I actually like their plan a little better than the ZeroTraffic idea. Whereas ZeroTraffic claims that they will solve traffic, DOVU sticks to the more prosaic claim that you can get paid to reveal your data. It could be in the future that cellphones will no longer report faithfully back to their corporate sponsors, perhaps because anonymous, blockchain-based payment systems have made it possible to conduct a phone conversation or data transaction without needing to keep tracking the mobile device. Or perhaps because some other bit of information will become as valuable as what DOVU believes transportation data bits are now. I’m not a fan of their tag line “blockchain powered mobility” as they do not directly provide mobility at all, but in the future DOVU could pivot based on an opportunity that presents itself and they could have a winning proposition. In contrast, ZeroTraffic is stuck on its path once it has sold its “deputy mayor of traffic” positions.

Me me me

My interests in applying blockchain techniques to transportation are based more in transportation economics than in engineering or computer science, and I haven’t yet seen anybody trying to implement my thoughts, so there’s still hope for my startup idea. Unfortunately, my idea will likely end up being an over-planned scheme like ZeroTraffic, and less a generic enabling technology like DOVU. Without giving anything away, I’m interested in dealing with that marginal cost pricing problem. I’m not going to blab about my ideas until I get more clue about blockchain and Etherium and all the other related tech. Maybe in a few months I’ll float my own ICO with my own team, and face my own round of critical blog posts pointing out how clueless my ideas are.

PS, If anybody from DOVU reads this and is interested in collaborating, shoot me a message and I’ll share more of my ideas. Or just recruit Professor John Polak from Imperial College London for your advisory board.

Writing tests and learning promises in node.js (part 2)

(In which I explore how errors are thrown and caught from within promises.)

Following up on my last post documenting work fixing up my bulkdoc appender code, for the paste few days I’ve worked on my couch_set_state code. The purpose of this package is to facilitate using CouchDB as a way to save state. I find this incredibly convenient for tracking jobs across multiple machines. I just set up my state tracking database to replicate between all my computers, and then can coordinate work between them by having code check the local CouchDB for state—things like whether a chunk of data has been processed, whether figures have been rendered, and so on. This package has utilities to set state, and a companion package couch_check_state facilitates checking state.

One of the things I wrote for the bulkdocs package was the ability to generate database conflicts. One of the design features of CouchDB is that it refuses to save a document unless the saver is updating the latest version of the document. The way my libraries work (both the bulkdocs appender and this state setting code) they first get the latest version of the document from the database, then they make their changes, then they save the updated document to the database. The problem is that if many jobs are running at once it is always possible that a race condition will arise where two processes are trying to update the same document at the same time. I actually hit this in practice last week, which is why I decided to write test code to trigger this issue. Continue reading

Writing tests and learning promises in node.js

(In which I describe how I did a naive transition to using promises instead of callbacks, and then did something slightly better.)

At the moment, I’m in an under-employed state. Hopefully the consulting will pick up soon, but instead of looking for a real job I’m cleaning up lots of old library code I use (and learning new things like Elixir and Kafka, but that’s a different topic). This post is the second about my work learning to use promises and async/await.

Today I worked on upgrading my code for saving and/or updating bulk docs to CouchDB (couchdb_bulkdoc_appender). As I wrote in my last post, I’m switching from mocha to tap, trying never to use lodash and d3-queue, and trying to use promises or async/await rather than callbacks. Replacing callbacks with promises is tricky because most of my code follows the usual node.js callback idiom. The last thing I want to do is break old code as a result of a stray npm install.

One strategy I’ve been using to migrate to promises is to test the callback function. If it is falsy (null or undefined), then I return a Promise, and if it is true, I run through my usual callback approach. For example, my old code did this:

Continue reading

Mixing async/await, promises, and callbacks

I’ve been editing some older code, but at the same time I’ve been trying to be strict about upgrading my node.js to a more modern idiom. That means wherever and whenever possible, I want to use node tap for my tests, and async/await and Promise instead of callbacks.

In the past, I have used the async library, and then later the d3 queue library for handling asynchronous functions. But no longer—my intent is to never use those libraries again.

Of course one big barrier to all this is that I have little experience with using Promises and async/await. This is the first in hopefully a set of posts that talk about some small things that I discover. Hopefully they will be useful to someone else.

First up, mixing callbacks and promises. As I said above, I’m switching to using the node tap library for testing. The docs indicate that the callback function passed into t.test (the signature is t.test([name], [options], [function])) should return a Promise in order to inform tap that this test is asynchronous. I’m writing new tests for older code, and I don’t have the luxury of rewriting my old code to switch from callbacks to Promises. On the other hand, I didn’t want to mess around with figuring out how to use old-style node.js callback conventions with t.test.

As an example, an old library code looked like:

// pass in year, config_file (optional) in opts object
function get_wim_imputed_status(opts,cb){
    var year = +opts.year
    var o = set_couchdb_options(opts)
    o['view']= '_design/wim/_view/imputed_status'
    o['startkey'] = [year]
    o['endkey'] = [year,"\ufff0"] // verified high sentinel as of 1.6.2 couchdb
    o['reduce'] = false
                  throw new Error('oops')
               return null
    return null

Pretty basic function—all it does is poke CouchDB and get some response. My old tests, using mocha, looked something like this:

    it('should get the imputed status of all the wim sites 2012',function(done){
                                             ... other stuff ...
                                             return done()

Eventually I will rewrite everything to use Promises, doing that now would also require rewriting my viewer library, and then rewriting all of the code that uses the viewer library, and so on. Someday maybe, but not today. So what I really need is a way to quickly wrap this callback in a Promise.

There are a few libraries that exist to wrap callback style functions in Promises. Node version 8.0.0 adds a very handy function called util.promisify (source code link is here), but I’m only running node version 8 on my laptop and on Travis CI, so that won’t work for me.

While the node.js Promisify wrapper is robust and covers lots of edge cases, it is pretty easy to hack up a much simpler version for more limited use. What I did was Continue reading

Learning Elm

After trying to learn Clojure and ClojureScript, I got to the point where I was able to make my map app and load it with data. However, loading the initial map of California took 10 seconds, and loading up and coloring traffic volume tiles on the map took another 5. Super slow.

This is not a knock against ClojureScript, etc, but rather a knock on my understanding and abilities in said language. I suck at clojure.

So I switched to Elm.

Here is my first trial with Elm, doing the letterfall thing.

Transfer prints

It is possible to make transfer prints from ink jet print outs. Professor Gerald R. Van Hecke was absolutely correct when he said, back in 1989, that I should use my knowledge of chemistry rather than saying lighter fluid “magically” lifts off the images from magazines. Had I listened, I might have been open to other methods.

Apparently, the techniques are all dependent upon chemistry—something needs to attack the bonds between the ink particles and the paper. In my old use of Zippo fluid and magazines, the lighter fluid did the trick. With Polaroid type 669 transfers, in the one case the ink hasn’t yet transferred to the photo paper, and in the emulsion transfer technique, the hot water dissolves the bond between the emulsion and the paper. These new-to-me techniques (it seems most articles on the internet were from 2011 through 2013, with nothing much new happening that I can find) some substance is used to lift the ink.

A good series of articles is here, a long article covering lots of different lifting media is here, and some all-in-one PDFs are here for gel printing and here for direct transfers. This last recipe is one of many approaches that print to non-porous surfaces (cheap plastic overheads; glossy backing to printable stickers; etc.) and then slap that surface down on the receiving surface before the ink has had much chance to dry.

So next weekend’s project is lined up I guess.

How to use npm to conditionally download a file

I am working on a package to process OpenStreetMap data, cleaning up some older code. My old REAMDE used to say "download the file from…", but sometimes people have trouble with that step. What I wanted to do was to automate the process, to check if the downloaded OSM file was older than some period (say 30 days), and if so, to download a new snapshot. I also wanted to use npm, because then it would cooperate nicely with all my other crazy uses of npm, such as building and packaging R scripts. Because I couldn’t find any exact recipes on the internet, here’s how I did it.

First, check out how to use npm as a build tool and the more recent why npm scripts. Both of these posts are excellent introductions to using npm scripts.

For my problem, there are two basic tasks I need to solve with npm scripts. First, I need to be able to check the age of a file, and second I need to be able to download a file. Note that because I only run Linux, I’m not even going to pretend that my solution is portable. Mac OSX users can probably use similar commands, but Windows users are likely going to have to change things around a bit. With that Linux-centric caveat aside, here is how I solved this problem.

File age

To determine if a file is too old I can use find.

find . -name "thefilename" -mtime +30

This will find a file called "thefilename" if it is older than 30 days (more or less…there is some gray area about how fractional days get counted). Rather than using this as an if statement, it’s probably easier to just use the built-in "-delete" operator in find to remove any file older than 30 days.

Download a file

To download a file, I can use curl. Specifically, I want to download the California latest file from geofabrik, so I would use

curl > california-latest.osm.pbf

Fitting into run scripts, mistakes and all

Delete the old file

My idea was to use the find command to delete a file that is older than my desired age, and then to use the curl command to download a file if and only if it doesn’t yet exist.

First, I codified the delete operation into a run script as follows:

"build:pbfclean":"find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete"

Running that failed spectacularly!

james@emma osm_load[master]$ npm run build:pbfclean

> osm_load@1.0.0 build:pbfclean /home/james/repos/jem/calvad/sqitch_packages/osm_load
> find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete

find: `./binaries': No such file or directory

npm ERR! Linux 4.4.10
npm ERR! argv "/usr/local/bin/node" "/usr/local/bin/npm" "run" "build:pbfclean"
npm ERR! node v6.2.0
npm ERR! npm  v3.8.9
npm ERR! osm_load@1.0.0 build:pbfclean: `find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete`
npm ERR! Exit status 1
npm ERR!
npm ERR! Failed at the osm_load@1.0.0 build:pbfclean script 'find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete'.
npm ERR! Make sure you have the latest version of node.js and npm installed.
npm ERR! If you do, this is most likely a problem with the osm_load package,
npm ERR! not with npm itself.

The problem is that find failed because I hadn’t created the destination directory yet. I don’t really want to create a directory just to empty it, so instead I tried running a test first.

So I extended the script a little bit:

"build:pbfclean":"test -d binaries && test -d osm && find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete"

This was another crashing failure:

james@emma osm_load[master]$ npm run build:pbfclean

> osm_load@1.0.0 build:pbfclean /home/james/repos/jem/calvad/sqitch_packages/osm_load
> test -d binaries && test -d osm && find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete

npm ERR! Linux 4.4.10
npm ERR! argv "/usr/local/bin/node" "/usr/local/bin/npm" "run" "build:pbfclean"
npm ERR! node v6.2.0
npm ERR! npm  v3.8.9
npm ERR! osm_load@1.0.0 build:pbfclean: `test -d binaries && test -d osm && find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete`
npm ERR! Exit status 1
npm ERR!
npm ERR! Failed at the osm_load@1.0.0 build:pbfclean script 'test -d binaries && test -d osm && find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete'.
npm ERR! Make sure you have the latest version of node.js and npm installed.
npm ERR! If you do, this is most likely a problem with the osm_load package,

The problem here is that the test -d binaries was doing its job, but was exiting with a non-zero condition. Reading the docs (npm help scripts) shows that non-zero exit is interpreted as a problem:


  • Don’t exit with a non-zero error code unless you really mean it. Except for uninstall scripts, this will cause the npm action to fail, and potentially be rolled back. If the failure is minor or only will prevent some optional features, then it’s better to just print a warning and exit successfully.

So test is the wrong tool to use here, so I switched to if;then;fi:

"build:pbfclean":"if [ -d binaries -a -d binaries/osm ]; then find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete; fi",

And the results are better:

james@emma osm_load[master]$ npm run build:pbfclean

> osm_load@1.0.0 build:pbfclean /home/james/repos/jem/calvad/sqitch_packages/osm_load
> if [ -d binaries -a -d binaries/osm ]; then find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete; fi

Unfortunately, while that doesn’t crash, I also want to check that it works to delete a file older than 30 days. So I made the directory in question, grabbed any old file older than 30 days, copied it into place and renamed it california-latest.osm.pbf:

find ~ -maxdepth 1 -mtime +30


james@emma osm_load[master]$ ls -lrt ~/3.7.10.generic.config
-rw-r--r-- 1 james users 129512 Sep 27  2014 /home/james/3.7.10.generic.config
james@emma osm_load[master]$ mkdir binaries/osm -p
james@emma osm_load[master]$ rsync -a ~/3.7.10.generic.config binaries/osm/california-latest.osm.pbf
james@emma osm_load[master]$ ls -lrt binaries/osm/
total 128
-rw-r--r-- 1 james users 129512 Sep 27  2014 california-latest.osm.pbf
james@emma osm_load[master]$  find ./binaries/osm -name california-latest.osm.pbf -mtime +30

Now running my build:pbfclean should delete that file:

james@emma osm_load[master]$ npm run build:pbfclean

> osm_load@1.0.0 build:pbfclean /home/james/repos/jem/calvad/sqitch_packages/osm_load
> if [ -d binaries -a -d binaries/osm ]; then find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete; fi

james@emma osm_load[master]$ ls -lrt binaries/osm/
total 0


Download a new file

To download a new file I need to run a simple curl command, but I also need to do two other things first. I need to make sure first that the destination directory is there, and second that the file does not already exist.

To make sure the destination directory exists, all I have to do is run mkdir -p. Alternately, I could check if the directories exist, and then run mkdir -p if they don’t, but that seems excessive for a simple two level path.

"build:pbfdir":"mkdir -p binaries/osm",

To test if the file exists already (and so to skip the download), I used if; then; fi again (having already been burned by test) as follows:

"build:pbfget":"if [ ! -e binaries/osm/california-latest.osm.pbf ]; then curl -o binaries/osm/california-latest.osm.pbf; fi "

Here the -e option checks if the file exists, and if it does not (the ! modifier before the -e) then it will run the curl download. If the file does exist, then nothing will happen.

Putting them together, I first call the build:pbfdir script, and then do the curl download check and execute:

"build:pbfdir":"mkdir -p binaries/osm",
"build:pbfget":"npm run build:pbfdir -s && if [ ! -e binaries/osm/california-latest.osm.pbf ]; then curl -o binaries/osm/california-latest.osm.pbf; fi "

(I couldn’t find the -s option documented anywhere in the npm docs, but apparently it suppresses output.)

It works fine:

james@emma osm_load[master]$ npm run build:pbfget

> osm_load@1.0.0 build:pbfget /home/james/repos/jem/calvad/sqitch_packages/osm_load
> npm run build:pbfdir -s && if [ ! -e binaries/osm/california-latest.osm.pbf ]; then curl -o binaries/osm/california-latest.osm.pbf; fi

  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0  578M    0  1194    0     0   1329      0   5d 06h --:--:--   5d 06h  1669^C

Of course, I could have just slotted the mkdir -p command inside of the build:pbfget command, but this is a better example of how to cascade two run scripts. And besides, maybe in the future I will be using a symbolic link pointing into a big disk, and so mkdir -p would be inappropriate.

The final scripts portion of my package looks like this:

  "name": "osm_load",
  "version": "1.0.0",
  "description": "Load OSM data (California) into a local database",
  "main": "load.js",
  "scripts": {
      "test": "tap test/*.js",
      "build": "npm run build:pbfclean  && npm run build:pbfget",
      "build:pbfclean":"if [ -d binaries -a -d binaries/osm ]; then find ./binaries/osm -name california-latest.osm.pbf -mtime +30 -delete; fi",
      "build:pbfdir":"mkdir -p binaries/osm",
      "build:pbfget":"npm run build:pbfdir -s && if [ ! -e binaries/osm/california-latest.osm.pbf ]; then curl -o binaries/osm/california-latest.osm.pbf; fi "