dormando.bsky.social

My home router were low power OpenBSD boxes for many years... Finally missing it. PF was super easy to manage with the NAT/FW rules being merged. Regular OS updates, minimal install, etc. Gave it up because UPnP wasn't working and games stopped working. Also required dedicated AP's.

"just use a router dingus lol" - yeah idk. I can't VLAN stuff on the wifi and that's the biggest issue. Otherwise it's suepr nice to be able to swap out the AP's without touching the router. Can still do transparent VPN's and other screw-ery. Nothing seems to do what I want?

My home router were low power OpenBSD boxes for many years... Finally missing it. PF was super easy to manage with the NAT/FW rules being merged. Regular OS updates, minimal install, etc. Gave it up because UPnP wasn't working and games stopped working. Also required dedicated AP's.

Finally got a report. I'm gonna "leak" it because the gymnastics are intense: - Enable TLS on a port with "bare" TLS (no cert auth) - Add a second port with mTLS - Enable server side ssl session cache - Connect to TLS port - Resume the session on the mTLS port - OMG cert auth bypass!

Casually skimming 'Designing Data Intensive App. 2nd ed.` for fun and refresh on stuff I don't use directly. Anyone who worked in social media should recognize the dreaded "timeline query" from an early example in the book. It then talks about an inbox-style TL + merges for famous folks. Rambling:

Since global post rate was "low" frequent refreshing would do much lighter queries. A real system would've had a service instead of BDB's in a MySQL column but it doesn't matter since I didn't finish it.

Casually skimming 'Designing Data Intensive App. 2nd ed.` for fun and refresh on stuff I don't use directly. Anyone who worked in social media should recognize the dreaded "timeline query" from an early example in the book. It then talks about an inbox-style TL + merges for famous folks. Rambling:

Of course we had the same TL query at LiveJournal. I recall making an awful prototype where I materialized timelines into a BDB blob and shoved it back into MySQL (lol). @bradfitz.com rightly told me to try harder (and I didn't, because I'm a fool). But there was something odd about it? Not like TW.

Casually skimming 'Designing Data Intensive App. 2nd ed.` for fun and refresh on stuff I don't use directly. Anyone who worked in social media should recognize the dreaded "timeline query" from an early example in the book. It then talks about an inbox-style TL + merges for famous folks. Rambling:

Was at TW office ages ago (some conference thing?) and someone was telling us about how many machines are just for delivering tweets for famous folks. I asked about delivering folks over N follows to a service then late merging with user timeline. "... yeah... we should do that."

So I kinda settled into not using AI for much despite what seems like everyone I've ever known going full tilt into it. Every 6mo I "update my skills" since FOMO and "left behind" is the big hype and... it just takes a day to get caught up. Any weird tricks are soon folded into the harness.

This doesn't mean it can't have _personal value_. I'd say strong parallels to 3D printing: there are no "3D printer user billionaires". I can print a command hook at home and not pay 3m $0.75 for a single hook. I'm never going to make a valuable competitor for command hooks.

So I kinda settled into not using AI for much despite what seems like everyone I've ever known going full tilt into it. Every 6mo I "update my skills" since FOMO and "left behind" is the big hype and... it just takes a day to get caught up. Any weird tricks are soon folded into the harness.

Not really needing it + the systemic obnoxiousness means I can happily minimize its use or ignore it. I'm not going to bother anyone else who does; this is just me saying I've done the math for myself.

So I kinda settled into not using AI for much despite what seems like everyone I've ever known going full tilt into it. Every 6mo I "update my skills" since FOMO and "left behind" is the big hype and... it just takes a day to get caught up. Any weird tricks are soon folded into the harness.

Part of this is the nature of my work: I've always done shit everyone else hates (automate things, figure out how things work, maintaining stuff) and AI isn't much of a shortcut there unless you like boomerang work. It's best at generating new junk and charging you for good and bad.

Story time: So I'm watching non-stop, because apparently I'd watch Liam Neeson take a shit, and there's a scene where the passengers are watching their own hijacking on the news. My wife says "that seems really weird". Then I'm thinking... wait, that seems really familiar? That kinda happened to me?

Not really, but: en.wikipedia.org/wiki/JetBlue... - This plane took off from LA to JFK at the same time as I took off, from JFK toward SJC. Both JetBlue flights. I think also the same plane? Our plane watched them circle for hours on the seat monitors.

Since Opus 5, claude code has been occasionally printing random chinese characters in its output. This morning I got a huge dump: half a line from my stuff then screenfull of someone's javascript UI tasklist in chinese. I imagine someone else must be getting random globs of my session output? No?

In b4: I don't often use this stuff. I'm positive it's not something weird I did. This is a CC session running in a dedicated container on a host I personally own. The only software installed is CC and various build deps. None of the text or code exists on disk. Happens more in long runs.

20 days, no security reports. will we make 30?

github.com/memcached/me... - Spent all of last week fixing every bug I could find. It's only been a few days since then but all the reports have been dupes or junk. Any new robots overstating a report will get banned without notice.

Only major takeaway is almost none of these were surprises. Nearly every bug had a nearby commit message or comment "there's likely a bug here" or "need to finish this refactor" or was a command with no tests at all. Years of poor follow-through on random subsystems due to working alone.

github.com/memcached/me... - Spent all of last week fixing every bug I could find. It's only been a few days since then but all the reports have been dupes or junk. Any new robots overstating a report will get banned without notice.

four more security bugs. monday monday monday m0ndae maendoea0.

wish I could post and/or blog through the development of this engine... but it doesn't feel... safe? anymore? just have to wait until it's done before getting into real detail.