Notices by Chris Siebenmann (cks@mastodon.social), page 2

@gnomon Embrace the world of platform pedals and regular footwear, I say from over here in the weird people's corner.

(People in my bike club periodically have a "you're biking in ordinary sandals?" double take to my footwear choices, which is always funny to me.)

Is it useful for a Unix sysadmin to know how to use ed(1) and do basic things with it? Sure. Sometimes ed(1) is all you have or is your best choice. Is it a priority for a sysadmin to learn ed(1), given that there are so many other things to learn in this field? I don't know, but I feel much less certain. There's a ton of things in system administration that are 'useful' to know, too many for any one person to actually know.

Sometimes I stumble over absolutely classical bug reports. Glibc bug report, 'glibc doesn't work right when filesystems have 0 inodes in directories'. Opened 2010, closed 2010 WONTFIX by who you think, reopened in 2016 under a new maintainer, fixed 2022-09 for readdir() and 2024-09 for readdir64_r().

https://sourceware.org/bugzilla/show_bug.cgi?id=12165

I discovered this via this Go bug about the same issue, https://github.com/golang/go/issues/76428 (which has a 'how you can get this yourself' example due to gvfs/FUSE).

@gumnos @rk As a sysadmin, I see and feel the appeal behind fast start/quick editing, but I've found that ed doesn't work for me because (I think) it requires me to keep too much context in my head. Visual editors give me visible context for me to readily work in, and with a stock/minimal vim setup they start and operate fast enough for me.

(I have a big, slow start editor environment for big editing. That's GNU Emacs, not vim.)

@whitequark I'm possibly a bit confused. A flock() lock is automatically released when the last copy of its file descriptor (its 'open file', the shared-over-dup/fork underlying fds) is closed. If you flock() and close the lock is gone.

If you do two open()s of the same file in the same process, flock() one (holding the fd open), and then try to flock() the other, AFAIK the second flock() will block, just as it would in a separate process. But I'd test it to be sure.

@whitequark I have some experience and investigation on NFS, due to work's NFS fileservers and us delivering email over NFS (so we really care about locking and I've had to pry up the floorboards a few time).

@lanodan @drscriptt example.org is a guaranteed placeholder domain (as is .com and .net). I'm not sure there's any other guaranteed ones and example.org is obvious and known to people; anyone with a SOA of 'example.org' or a CNAME to it is saying 'this is not real' in an obvious way.

@drscriptt Roughly, $TTL very large, SOA claiming everything at example.org with very large timers, NS to the nameservers of the particular registrar's nameservers (could have been 127.0.0.1), MX ., A to 127.0.0.1. An alternate would be a wildcard CNAME to example.org, since that has much the same for eg MX.

CNAME to example.org avoids any 'you people are copying my content/hacking my network!' rants that might somehow wind up reaching us.

Have I just written up a simple poison zonefile that should make everyone querying a zone (that we don't serve) go away for as long as possible? Maybe I have. If I have, it's certainly not deployed in production. Not yet, at any rate.

If you list random DNS servers as the servers for your domain, you do not get any quality of non-service guarantees.

Current status: enjoying the use of PF's "probability ..." option to divert a portion of traffic to a new server under test. Also, discovering things under rocks that have been turned over by tcpdump'ing DNS traffic to a new authoritative server under test.

Are there people out there in the world who apparently blindly list three UofT DNS servers as their authoritative servers (when the servers aren't)? Apparently yes, and all from one registrar so far, so it may be Registrar Hijinks.

@ireneista Working in/for a Computer Science (university) department has been a useful thing for me for this nuanced view of 'AI'. While there''s an overall 'AI' supergroup in the department, it has a bunch of named sub-groups called various things (local names and abbreviations include CL, KR, ML, and Vision). When we routinely have to ask 'okay what *sort* of AI is that new professor associated with?', it sticks in one's mind.

(And they're not all one big happy resource-sharing family.)

I have opinions about Go's forced switch to modules, and one of them is that it effectively destroyed one part of the Go 1 compatibility promise in practice.

I can take Go code from 2015 or 2019 and run it today. Sure, great. But I generally can't take a Go program from 2015 or 2019 and rebuild it today, not easily, because you can no longer build non-modular programs. Converting a Go program to modules is non-trivial, especially third party programs.

Apparently RAM prices are going through the roof, both DDR5 and earlier DDRs, because thanks LLMs, and they're expected to stay that way for maybe all of 2026¹. I guess my 2018 home desktop gets to stay being my desktop for another year or two.

(It's not like it's suffering from performance issues, although it does now take 45+ minutes to build Firefox.)

¹ Oh look, another reason to hope for an LLM crash that guts the demand for new GPUs with lots of RAM.

I did not have my commuter bicycle converting itself into a 40x15 singlespeed on my bingo card for today, but fortunately it didn't go for the 40x12 option and the local bike shop is on my way home and not too far from home (and open at the time).

More or less getting up the St Clair hill was interesting. And somewhat sweaty. I do not have @mhoye's mighty thews, to put it one way. (I am a gearing wimp, people who ride singlespeed up hills are impressive but I will watch from a safe distance.)

@whitequark @glyph @dotstdy @becomethewaifu @mcc As Catherine knows, "Straight up delete null pointer checks" isn't a hypothetical example: it caused a LInux kernel security issue in 2.6.30. The full gory code can be found in one of the examples in https://wiki.sei.cmu.edu/confluence/display/c/EXP34-C.+Do+not+dereference+null+pointers and a summarized version in https://utcc.utoronto.ca/~cks/space/blog/programming/CUndefinedDereference

@evan I voted "yes, but" on the grounds that reverse engineering how the dishes are so good may cause you to learn things you did not want to know about their preparation, like (allegedly) just how much butter goes into any number of them. Ignorance can be bliss, or lead to blissful continued consumption of restaurant dishes.

But maybe you wanted to know anyway. Reverse engineer away.

Against my better judgement I've just emailed feeder.co about their persistent habit of sending a 'bypass_varnish=1' HTTP cookie on all feed fetch requests. It does nothing positive for them with my blog, but it's a hostile thing and I'm no longer interested in tolerating hostile things from people hitting my blog, so if they don't fix it I'm going to block them in a week or two.

@marginalia My notes say the two incidents were 2025-03-23 and 2025-09-01, I believe both times shortly after midnight US Eastern time. This was for my techblog, https://utcc.utoronto.ca/~cks/space/blog/

As far as the UA: it's standard for crawlers to include at least a http/https URL that explains what the purpose of the crawler is, how to contact people related to it, any ways to get it to slow down such as special HTTP headers in responses, etc etc. 'search.marginalia.nu' is ... minimal.

Sysadmin hive mind: are there any Unix/firewall/etc security related software that are worth paying for (apart from Tailscale)? The kind of thing that you'd look forward to your work acquiring, instead of dreading it.

(Assume that MFA is covered.)

@lanodan @mirabilos I suspect font weirdness around some sort of package update, since I did do a package update and I had a brief thing where a new xterm window came up with odd fonts (I closed it immediately and didn't look further).