Mark Tomczak @mtomczak@qoto.org

@brett I'm sure it's just the confirmation bias talking, but I can't keep track of how many times I've seen a story about "Massive exploit in some proprietary special-purpose software that definitely has an open-source-based alternative leads to lost data / ransomware'd machines in $BIGCORP and $BIGGOV sector."

In general, I appreciate APIs where `{foo: null}`, `{foo: []}`, and `{}` are not only equivalent but indistinguishable from each other.

Otherwise, you're giving me an urn with a scorpion in it and I never appreciate that.

@Blue_Jersey Depending on your use case, Git gists (https://gist.github.com/) are a tool some people swear by. Git will do a pretty decent job of auto-formatting the content if it knows about the language.

@rauschma I'm actually using a server that supports Markdown. Details of how / what would need to be done to support that on someone's server, I do not know.

#cplusplus #rust

You know that thing about "You can't look up a recipe online without having to scroll to the bottom to get past the writer's life story?"

The software equivalent of that is looking up anything about commercial software systems. You Google them and then you have to scroll down past all the vendor's advertisements and their main page where they tell you "MoonClover is a rethinking of the nature of the nature of knowledge, the next step in evolution of the human comprehension of reality. The top choice of forward-thinking tech firms, it's pushing through ignorance to shine new light on the nature of existance and..."

Fuck that. You scroll down to the Wikipedia article so you can read "MoonClover is a key-value store with performance equivalent to MongoDB that was created by Bob Bobson and Steve Stevens to get their company, BoboStevo Inc., off paying the Oracle DB tax every quarter."

@edporras This is why my blog's in Hugo.

Never have to worry about CORS or XSRF attacks when there are no JavaScript callouts on the page.

@randomgeek I'm really trying to wrap my head around this "search engines are useless" meme.

Like, it's a thing people keep saying, but then I go search for something and get a useful result. Consistently, four-to-five-nines percent of the time.

🤷

@mo8it Some of the time. We have a huge codebase in a monorepo managed by Bazel rules and it's real easy to push the language server right over trying to understand the whole thing.

I'd say it works about 60% of the time.

@mo8it I believe that feature exists in C++.

... and I'm pretty sure I hate it. It's spooky action at a distance. "Why the hell is this code behaving this way. I'm looking at the trait implementation, it's nothing like what the code is doing... *Oh I see there's another implementation of the trait for a concrete template specialization all the way over here in a different file well I guess my time is worthless then.*

(Note: this wouldn't actually be so bad if C++ had a way to say "For this variable, what function is being called from this line?" But it doesn't, so specialized template implementation suck on toast for discoverability).

@argentcorvid Oh. I guess I'm just having my searches successfully satisfied by SEO spam then.

@rauschma Those are default names Reddit assigns when you don't pick a name.

It's the laziest spam approach ever and it still works.

@catvalente Were you able to get a non-smart TV? I'm terrified that the next time I have to replace my ailing 14-year-old TV I'm going to be stuck with something that's not a TV, but a dedicated-purpose computer that happens to have "TV" as an application it can run (and all the irritations that comes with... Literally had a relative where I had to drive over to their house to reboot their TV once a month because the damn thing would just freeze).

@eh FWIW, the problem you're hitting is challenging because it *is* hard to get both performance and correctness with threads. Most runtimes I'm using day-to-day "solve" this by functionally banning threads and making execution fully-deterministic (and, as consequence, sacrificing the potential performance multiprocessing allows).

If we want multiprocessing, we bundle the behavior into an entire queryable service and run it on multiple nodes. This is expensive both in terms of code labor (i.e. switching from a subroutine to an entire Dockerized service is more than a couple hours work) and overhead (now there's a whole IPC framework in the loop, if not a network fabric).

I hear rumor Erlang simplifies this but I've never looked into Erlang closely enough to really know.

@zoomar Can you give an example query? I've definitely heard people say that most of their search results are Reddit these days, but I almost never get Reddit; I assume I'm searching in domains where Reddit hasn't dominated the index and I'd be curious to know where it is.

@urusan At Google, heroics were an auto-trigger for a postmortem (or at least, they were on solid, well-functioning teams).

Something came up that made you leave the office late? Carve out time in the morning to do at least a memo to the team, if not a full post-mortem.

#reddit is foundering today.

I'll be interested to find out (if I ever do) why the Reddit Strike appears correlated with general service degradation. My guess is that either trying to access a private subreddit is actually an expensive query and Reddit never realized that because they never had to test that interaction at scale, or tens of thousands of bots that treat Reddit as infrastructure and auto-archive or copy the most popular subreddits are now finding their access cut and their recovery protocol is busted, and Reddit never had to test denying access to so many bots (probably not using the API, probably scraping the page and faking being real users) at scale before.

@aberrant68 Thank you. I appreciate this.