There is no such thing as reliably cleaning up after a test failure. If you could do things reliably, the test wouldn't have failed.
I've been travelling in Canada and wondered why my phone's data plan was so fast. Turns out that sometime in the last year, T-Mobile improved their international roaming to run at full rate (instead of intentionally degraded 128kbps) in Canada and Mexico, plus added free unlimited calling from those countries, just like in the US. Okay then.
I'm still a little squinty eyed at "binge on" but even with that violation of strict net neutrality, I haven't been able to figure out a way it's bad for consumers, and it's a massive aid when it comes to Digital Divide.
I am unreasonably pleased with this service.
Unfinished/unreleased software changes are inventory. Inventory is a cost without a benefit.
From Deming's "Out of the Crisis", talking about Americans' experience visiting Japanese auto plants in the 1980s:
"Shipments of stampings, subassemblies and assemblies are made directly to the auto assembly line several times a day. Side-loading trucks drive into the assembly plant and containers of parts are dropped off at the appropriate work-station on the line without incoming inspection and counting. Parts are assembled into the automobiles, as received. The virtual absence of inventory results in an estimated space saving of 30 per cent as compared with comparable U.S. auto operations."
Avery's flu-induced theory of the day:
C is like cvs. Old, hacky, but fairly easy to understand, available everywhere, and you can fix it when it breaks. Also, a little too much dependency on arbitrary text substitutions in your source code.
Go is like svn. Nice clean design, intended for users of the old system to migrate fairly easily, but there are a few fundamental design decisions that just grate on you day after day. But businesses use it because it's solid engineering and you can pay people to put up with minor annoyances as long as the work gets done in the end.
Rust is like git. Totally bonkers, bare metal, user-hostile, and only a small few can understand it, but somehow it beat svn in the open source world because it let people solve problems that svn simply couldn't do. (Clearly Rust hasn't beat anybody yet, but this is my flu-induced theory so I can theorize whatever I want.)
Wow, that was unexpected.
I'm sitting in Old Montreal in the Crew Collective Cafe: https://crew.co/cafe/
The Yelp reviews for this place were emphatic and consistent about the "super fast wifi" among other things, so of course I was intrigued and came for a visit. (Turns out the coffee and pastries are also really good.)
The speed test results are quite unimpressive: 10 Mbps down, 5 Mbps up. The free wifi seems to be just a single dual-band router. On 5 GHz it's set to channel 36, 20 MHz wide, 802.11n, which is surprisingly narrow and slow, even though the other channels are mostly empty. But on the other hand, there are probably 100 people in here, everyone on a laptop. Unprompted, my friend said to me, "Wow, the wifi in here is really fast!" just based on browsing around on their laptop. So something is up.
My guess is there's just something really good about the airtime fairness on the AP in use. The giant open room with cement walls probably doesn't hurt either (my Macbook says the noise level is -99 dB, the lowest I've ever seen anywhere). This is more good evidence that raw speed is irrelevant to almost everyone (except the occasional large upload/download), but consistency and low latency are king.
Anyway, of course I was curious to know who makes wifi routers that are this awesome, so I thought, hey, I'll just look up the OUI (9e:15:54 and 82:15:54). Well, the joke's on me: both of those are not listed in the IEEE's OUI list, so something funny is going on.
The adventure continues...
As I continue to read all this stuff about Deming and quality control, I made a realization: Amazon Prime is just a natural outcome of quality improvement.
We keep imagining (and they want us to imagine) it as something where Amazon had to spend extra money to deliver your stuff in "only" two days. But the truth is, you really should be able to get most stuff from point A to point B in two days, as long as that stuff wasn't more than two days away in the first place. And Amazon has lots of warehouses scattered around the country, so it mostly isn't very far away. (And for items that are too far, they don't offer Amazon Prime!)
We've gotten used to the Post Office delivering stuff really slowly, then charging us extra for more speed. But the speed differences are mostly artificial, due to unnecessary slowness throughout the system. A natural consequence of fixing the system is that deliveries will happen sooner with less variability, and the delivering company will save money.
Yesterday I was having a random conversation with someone in who mentioned that he noticed, during an outage, that all our LEDs blinked in sync, and realized, "wow, someone had to have made this work on purpose." It was a good feeling. Sometimes the OCD pays off.
I think this is the sort of "I can tell someone actually paid attention here" that makes people like Apple products. Admittedly, it works better in cases where it's a little more useful than LEDs blinking in sync, but oh well. Baby steps!
Moved to new apartment. Listed old one for sublet. Got modified Nigerian Prince scam. That's a new one (to me).
- I'm a foreigner and my employer is paying me to relocate, price is no object!
- Okay, I guess that's not how I would have opened the negotiation, but no need to rip you off, what are the dates you're staying?
- I'm super nice and never make noise and am like the best tenant ever
- That's great, what dates?
- Could you tell me what international airport is closest to you?
- Um, it's New York, so...
- The thing is, my employer doesn't like to issue two cheques, so I need to get them to send you one cheque for the first month's rent as well as my travel expenses, and then you can pay me back for the travel expenses, okay?
It finally happened! I bought a new humidifier, which for no reason includes an IR remote control. That IR remote control uses IR codes which overlap with my (2011-era) AppleTV.
I knew overlapping IR codes happened, and have been frankly surprised that they have never happened to me in my lifetime, as far as I know. Well, lightning finally struck.
(Problem was resolved using masking tape, because I couldn't find my electrical tape. It took 4 layers to mostly block the signal, and 5 layers to fully block it. Apparently masking tape isn't so good at masking IR...)
Things get better when people are measuring them, and worse when people are not measuring them. To improve engineering culture, you have to make it easier for people to measure the right things.
Over the holidays, I read a lot about Deming, just-in-time inventory, and how if you have "multiple inspectors" of an output it works worse than one inspector, because all the inspectors will assume the other one has checked the work, and regardless, nobody wants to be the one inspector that makes the others look bad.
Compare and contrast with how I need 3 VPs, a manager, a director, a lawyer, and a finance person to sign off on a small contract to pay an open source developer for some work.