Show more

I used to overengineer, now I overpolish.

Should learn how to stop at "good enough".

Holy shit am I productive with and .

is for thinking and making sure that you're building the right it, is for doing and making sure you're building it right.

Welcome , y'all, your friendly hardener for our / implementation called .

asciinema.org/a/zCy4Ro5vPRPShg

We probably should set up monitoring, @pola ?

I have log aggregates set up for mail server, but I feel like that's not enough too...

Something for me to figure out after ? :)

Show thread

On the subject of re-writing the same thing over and over again.

For I need to capture detached signatures[1] in .

I made a `:> Capture ` query on website[2] and after very very brief exploration, an API capturing something called `SessionToken`[3] caught my eye.

Upon inspection it turned out that they have B64 encoding as well, except significantly less deranged than mine.

[1]: hackage-search.serokell.io/?q=+
[2]: hackage-search.serokell.io/vie

is very stable in terms of memory consumption!

But I'm thinking about what we're really doing here in and I'm comparing it with, let's say, resource footprint of community (Zaibatsu / ... / Soviet) and realise that their footprint is kind of zero.

Everyone is on one of the three manually federated servers, using UNIX / pre-UNIX tools to communicate with each other, with the only non-UNIX tool being IRC.

Somehow it doesn't take 2GiB of memory and somewhat constant CPU load for to exchange bytes.

I get that to host huge instances like mastodon.social, it makes sense to have , but it feels like individual instances should be able to be cohosted on the same machine.

I'm committed to explore using file system and files for persistent storage backend.

But that also brings me to a dilemma of whether or not to use for the rewrite or to use , to decrease footprint.

Obviously, will yield more stable, fault tolerant code with easier replication. , however will reduce excess footprints to 0 at a cost of longer development time (I'll need to benchmark current -based HTTP servers and see if I need to write my own dead simple server that doesn't use generics / isn't bloated), and it can easily eat up a month of my time.

Maybe it's all non-issue, especially if there's just one BEAM machine launched per host and we plug our applications into it (as intended)...

I think that the fact that I'm heavily leaning towards as the main production technology even though "close-to-metal" argument is important to me is a clear sign I should use it for production in .

Ended up with a pretty gross and not total From/To-JSON instance for Signature:

git.sr.ht/~jonn/do-auth/commit

I'm embracing partial functions on purpose (I wish had 's Partial typeclass).

I'd like my mission-critical software to crash when something unexpected happens and rather enter a recovery mode, a la

But that's also the reason why I'll most likely do a rewrite of in for v1 -- fault tolerance facilities are virtually limitless in the world.

Going to live-stream some implementation in . You can talk to me in this thread. :O

Doma Social

Mastodon server of https://doma.dev.