It took me about a month to get over the lack of “convience” of no cards & needing to carry cash in my country, but now I absolutely prefer it—as do small vendors. Card usage is very low & if you use one, minimums are high & you are on the hook for the network fee. Where I was born, folks bake the cost of the fee into the price which means using cash, you are subsidizing everyone using cards/apps which is pretty insane—it’s money for the giant payment providers that vacuum up all of our personal data. Cash is resilient, but way more private.

Snikket is meant to be super simple to self-host. Ejabberd has a web GUI that can make configuration easier.

History / sync is known as message archive management (MAM) & every normal modern client & server supports it. OMEMO uses same double-ratchet encryption & multiple clients as Matrix (with the same old client key dropping issues sadly). By default it does not support groups you are correct, however, FOSS Jitsi (& Zoom for that matter) is powered by XMPP under the hood & can be stood up by yourself.

Personally three of my circles have opted for separate Mumble servers for voice coms (I run one of them from my living room) as video is only ever rarely needed & the system resources is minimal. Having web cams on is seen as a chore & distraction sometimes. The only time video is helpful in my experience is screen share which is different—but screensharing is the worst tool for trying to do code pairing / debugging a terminal using upterm provides a crisper view experience, lower data/system requirements, & observers can optionally drive the remote session.

You could switch some of the problems with perf in switching away from the Python implementation server as well as Element clients but these support the most up-to-date features & the majority of users are now relying on these features that often don’t degrade graacefully.

The bigger issue is eventual consistency. Eventual consistency will not scale for small self-hosting. Every message & every attachment for every user in every chatroom they have joined must be duplicated to your server. This is why joining rooms sometmies takes 10 minutes. Even if you make this async from the client side instead of the current long wait, your server & storage are still taking the hit. A lot of small collectives had to drop their servers for performance & cost (read about yet another one today on the Techlore thread at c/privacy where now only Discord is used for realtime coms). This model is required to copycat the ability to search the entire history like the big, proprietary chat apps such as Slack/Telegram/Discord, but they are centralized so it is easier to manage—but its overuse for all announcement & trying to replace forums turns it into a black hole for information. Your small community probably does not need persistent chat like this—persistent info is lighter & easier to crawl as feeds & forums. With medium-sized servers shutting down, only the biggest & smallest hosts are still kicking with most metadata is largely centralized around Matrix.org who also hosts some of the other larger instances.

If you agree that chat can be chatter as well as ephemeral there is lightweight centralized chat in IRCv3 with TLS has most of the features you need with a longer legacy & massive choice for clients & XMPP for lightweight decentralized chat with a long legacy, client options too, & can be self-hosted in a bedroom on a toaster in comparison which increases the chances of self-hosters & decentralization. These were built in a time when we didn’t have such wasteful taste in tech since they needed to be efficient & only sip power/data in comparison both for clients & servers & storage. The bigger question IMO is what are fundamentally wrong with these two mature options that we need a new option built on unextensible JSON & Israeli Intelligence money?

Exactly. Our words matter & the sooner we stop using Google to mean search or MS GitHub to mean code, the sooner we can start shifting the narritive towards entities that better respect our privacy or even gasp self-hosting. Word choice for social change is just as important for spreading the message.

If you want a rush, try gyukuro. Flavor of an umami green tea with the caffeine level of a cup of coffee & high L-thiamine to cut the jitters.

I lived it the UK briefly & my roommate never understood my fascination generally preferring his Yorkshire tea (trash). One day he asked about it so I made him a mug (we used coffee mugs). He said it tasted like yard clippings like other green teas before walking off continuing to sip to not waste it. He comes back in 15–20 minutes later over halfway finished, “Did you put drugs in this‽”. “Nope.” “I feel like I want to run a marathon. Okay, I see the appeal for your mornings, but this is the worst high ever.”

Gerrit is probably the poster child for branchless, stack-based diffs in Git. It takes some get getting used to, but once adjust your thruput is really ramps up. In some sense tho, this is a hack by tagging changelist values in the commit message to help reconstruct what the heck is going on due to Git limitations, but it’s old & robust enough to trust that system & many of its users absolutely swear by it (I have limited exposure but have used it more recently I can feel the appeal). You should be able to slap it in front of any Git server—even just straight host HTTP if not something lightweight like cgit, gitweb, or Ayllu. (Jujutsu is the same commit hackery in a different package & I don’t think it moves the needle as much as folks think being ultimately shackled to Git’s design decisions).

If you look outside of snapshot-based tools like Git, Mercurial, & so on, patch-theory-based options offer refuge. Darcs & Pijul are the leading (D)VCSs in the space. Darcs is very mature & shows its age in many ways (but is still developed & works good enough). Pijul is largely based on Darcs but meant be faster (& is), but it is immature; some features are missing on purpose to avoid the swell of Git commands, but I am personally surprised theres no good story for sending patches nor rebase. That said its identity system is how VCS should do it. Both VCSs have a lot less tooling built around them. Darcs is still supported by tools like Nix (but not Flakes) as well as Opam for OCaml with darcs hub & Smederee for maintained public forges. Pijul isn’t supported by much at all unfortunately & while Nest is a public forge, its lacking in features & basic usability like being able to fetch a tarball (despite pijul archive). All the latter negativity may sound bad, but all tooling requires momentum. They would be prime candidates for the Gerrit workflow--just without the hacks needed. With the two being similar, I hope we see more tooling pop up to support them & just like trying a new paradigm of programming gives you insight on the ones you know, a new way to do VCS will teach you about version control. Do recommend.

client or server that doesn’t support the same encryption protocols

Outside of TLS which most any server uses by default, XMPP or not, the server is not responsible for E2EE. Conversations Compliance & Are We OMEMO Yet have existed for a long while & I never see anyone recommending a client not on these lists so while certain features may be fragmented, the communication essentials have been more or less established for years now. XMPP is an extensible format, and some applications that aren’t for chatting with your friends/family, don’t need many of these features which allows the protocol to morph into something stripped down for the task… which is why the base spec is basically barren, & community XEPs are what folks get behind for adding new features for different use cases.

That’s really cool to hear 😀

I do prepaid & get a new plan anually. It’s cheaper this way, less surprising bill-wise, & the rotation of numbers can help with security & privacy. All the carriers suck here equally but coverage is the same so I just see who is offering the best deal.

So is WhatsApp, Zoom, Jitsi

Meanwhile Matrix was built & funded by Israeli Intelligence (to which I’m sure there are anonymous donors today). It’s expensive replication model means only those with the deepest of pockets can run a server leading many to flock to the mother instance of Matrix.org centralizing, replicating the data to a single node (being decentralized in theory, not so much is practice). It’s funny to see them call out Signal, but luckily there are private, free alternatives to both.