[NOTE: The release of cider deprecates much of the content here. I will post an update on Clojure Debugging '14 early in the near year]
I’m ramping up for a new set of development projects in 2013 and 2014. My 2010 era setup with slime and swank-clojure is unlikely to remain a viable approach throughout the project. I’ve decided it is time to join the nREPL community as well as take advantage of some of architecture innovations there which may make it easier to debug the distributed systems I’m going to be working on.
Features I’m accustomed to from common lisp slime/swank:
- Code navigation via Meta-. and Meta-,
- Fuzzy completion in editor windows and the repl
- Documentation help in mini-buffer
- Object inspector. Ability to walk any value in the system
- Walkable backtraces with one-key navigation to offending source
- Evaluate an expression in a specific frame, inspect result
- Easy tracing of functions to the repl or a trace buffer (in emacs)
- Trigger a continuable backtrace via watchpoint or breakpoint
Only the first three of these features is available in the stock nrepl. The rest of this post will discuss how to setup a reasonable approximation to this feature set in Emacs using nREPL middleware providers as of May 2013.
The Fog of War is invoked to describe the uncertainty that permeates combat operations. This term is generally attributed to a quote by Clausewitz: “The great uncertainty of all data in war is a peculiar difficulty, because all action must, to a certain extent, be planned in a mere twilight, which in addition not infrequently — like the effect of a fog or moonshine — gives to things exaggerated dimensions and unnatural appearance.” (via wikipedia).
The evolution of an innovation or innovative company faces a similar fog. At any point in time, we extrapolate from current conditions a set of possible future outcomes and take action to try to bring the best outcome into being. In one view, we should make economically rational decisions, those that maximize our expected return over the outcomes weighted by their likelihood.
However, Nassim Nicholas Taleb’s “The Black Swan” argues convincingly us that the seeds of disruption and destruction lie in events that exist outside all reasonable extrapolations of a current state of affairs. The occurrence of a highly improbable event is, in fact, inevitable.
Since we can’t know anything for sure, innovators have to operate at a completely different level than analysts. It’s a chess game where the rules can change mid-play forcing a complete re-evaluation of strategy. For example, the iPhone was a Black Swan to the cellular industry, as this analyst’s perspective shows. This dynamnic is why you should ignore Forrester and other analysts – they tell you what the world will be like if nothing surprising happens, but surprising things always happen.
Recently I’ve been observing via RescueTime that I spend 3 hours or more hours in my e-mail application most days. However, I don’t have a good breakdown of how much of this is scheduling, looking up information, commenting on something substantive or social discourse. There is a tremendous amount of information locked up in the time-series of e-mail’s sent and received that can provide insight into aspects of my behavior such as focus of attention (time of day e-mail is sent), social relationships (what organizations I interact with in a given week), the density of idea generation, etc. E-mail logs contains a wealth of raw data that can be instrumented to uncover important information about our life.
Our E-mail logs are also rich archive of useful information such as phone numbers, addresses, what we said to someone, when we said something to someone, edits to papers, attachments, etc. With a proper set of tools, many of which have been built for analyzing social media, we can turn this archive into a database of useful information that can significantly enhance e-mail-based instrumentation.
My github fork of the Clojure library for HBase, clojure-hbase is now deprecated. I’ve extracted the functionality from David Santiago’s original library (with permission) along with a duplicate of his admin functions to create a parallel repository with the schema-oriented API I developed.
At my first startup company, Silicon Spice, we were extremely disciplined in how we structured internal communications, identified and followed technology standards and incorporated automatic testing into all our technology development efforts. The company-wide discipline we established allowed for an unprecedented pace of product innovation in the rapidly evolving communications market of the late 90’s. Peaking at only 120 employees, we constructed a complex set of technology targeting telco equipment vendors including reference voice switching boxes, a 2.5 Watt 21-core DSP processor, companion ASIC devices, a vector compiler, a real-time OS, a fully-featured set of voice processing software and a complete suite of development tools for our platform. A few people in key roles were able to spot opportunities that spanned numerous product sub-teams and we were able to quickly implement coordinated design changes across our home-grown technology stack.
A lead engineer from Texas Instruments, our primary competitor, once commented to me that they couldn’t believe that we had build our product in only 3-4 years with 100 people; I didn’t have the heart to tell him the product they were evaluating for acquisition was built by an average of 50 people over 16 months! We seized 70% of the carrier-class market from them in the 18 months following our acquisition.
I recently wrote a plugin in Clojure to add to the Cloudera Flume framework. As it was my first time writing a full java class interface I had to learn about the proper use of both proxy and gen-class. Given the poor error reporting at the java-clojure boundary, figuring out what you did wrong if you don’t get every detail exactly right (particularly when loading a class in the plugin’s final environment) can be difficult.
The clojure-hadoop library is an excellent facility for running Clojure functions within the Hadoop MapReduce framework. At Compass Labs we’ve been using its job abstraction for elements of our production flow and found a number of limitations that motivated us to implement extensions to the base library. We’ve promoted this for integration into a new release of clojure-hadoop which should issue shortly.
There are still some design and implementation warts in the current release which should be fixed by ourselves or other users in the coming weeks.
I have a bad tendency in my research work to write my own code and libraries from scratch, in large part because I’ve decided to keep most of my coding in Common Lisp to leverage prior tools. However, I’ve recently been given a painful demonstration of how it is often faster to pay the up-front cost to learn the right tool than to rewrite (and maintain) the subsets you think you need. For example, I found myself venturing into Clojure/Java/Hadoop for my commercial work this year as a compromise between Lisp / dynamic language features and integration benefits. This week I’m finding the need to do some rather sophisticated work with graphical models and I need some tools to build and evaluate them.
I’ve looked at a wide variety of open source approaches such as Samiam (no continuous variables), WinBUGS (only windows), OpenBUGS (not quite ready), HBC (inference only), Mallett (OK, but I don’t like Java and doesn’t support all forms of real-valued random variables), Incanter (limited but growing support for graphical models) and R.