Wilfred Hughes @wilfredh@mastodon.social

An introduction to the world's largest CRM, the importance of allowing users to build custom logic, and a discussion of providing a programmable AWS Lambda style platform: https://tryretool.com/blog/salesforce-for-engineers/

Playing with perf today! It's really interesting to see low-level details of where compute time is going. Branch prediction works well most of the time! (At least for this workload.)

Based on https://jvns.ca/blog/2014/05/13/profiling-with-perf/ and https://users.rust-lang.org/t/profiling-in-rust-application/18195/2

The dominance of statistical models in AI, our bias towards embedding human knowledge, and the effectiveness of large, generic compute:
http://www.incompleteideas.net/IncIdeas/BitterLesson.html

Other programmers using your code is often a high compliment.

Making something from scratch is fun, but learning someone else's API is work.

Rather than the common "I built one, you should use it", I try to ask "what would persuade you to use this?".

Scarcity versus abundance mindsets in software design, and shipping rough prototypes early: https://breakingsmart.com/en/season-1/rough-consensus-and-maximal-interestingness/

What's the threshold for politely pasting text in IRC channels? I pasted two lines and my client sternly warned me and mentioned pastebins.

This paper is from http://sunnyday.mit.edu/papers.html#ft, which includes some interesting follow-up work too.

This wasn't tied to experience either. The most experienced programmer had a program with some buggy outputs that overlapped with other implementations.

They had 27 programmers write the same radar analysis program, and then fed a million inputs to see if the programs differed.

For some inputs, up to 8 programs agreed on a wrong output! Having the programs vote could give you the wrong answer.

This is an approach used in some safety critical systems, but the paper finds that failures are not independent!

Programmers make similar mistakes, especially around boundary cases.

An Experimental Evaluation of the Assumption of Independence in Multi-Version Programming, by John Knight and Nancy Leveson http://sunnyday.mit.edu/papers/nver-tse.pdf

If you take bunch of programs written independently, are you more likely to reduce bugs by taking the most common output?

Excellent discussion of a Unison meetup, discussing their design (globally consistent content-addressed codebases!), type system, and tooling: http://unisonweb.org/2019-04-04/first-meetup.html#post-start

It's really impressive how much they've achieved.

Friday's xkcd is blunt but fair: it's very hard to secure any part of a modern computer stack, and we depend on all of it: https://m.xkcd.com/2166/

Apparently you can buy smart power sockets now, e.g. https://www.amazon.co.uk/Lightwave-L42-Smart-Socket-Stainess/dp/B075XT62RC

Whilst a consumer might enjoy the remote control, these devices also monitor power consumption. I imagine this would be particularly useful in a commercial property context.

Different computing platforms have different notions of 'rich text pasting'. Pasting styled text from a browser to a document editor is fraught with all sorts of compatibility challenges.

Is it worth it? Text formatting is very complex and representations depend on application.

It's remarkable how similar this is to the organisational approach I've seen at many tech firms! The problem solving, the lean team, and the bugs that can arise in that environment.

Quote from the article: "we always talked about buying ships and manning them with people … [it's different] now. We need to figure out how to have the fewest number of people possible, and then build [ships] to make them as effective as they need to be"

The US Navy is exploring ships that are more highly automated. The crew is smaller, and their roles are generalist problem solvers.

https://www.theatlantic.com/magazine/archive/2019/07/future-of-work-expertise-navy/590647/

Emacs trunk is now using HarfBuzz for text rendering! https://lists.gnu.org/archive/html/emacs-devel/2019-06/msg00123.html

This approach also finds error cases that crash the tool. If you get a traceback, its text is simply detected as not relevant!