bonifartius 𒂼𒄄 @bonifartius@qoto.org

Fedi ?

WaPo, a politically active publication owned by the third richest billionaire in the world, running articles warning about politically active billionaires will never stop being hilarious

Vietnamese dishes. Real local food. Cheaper Yummier. #foodie #vietnam

The strategy is this: move my pinky finger.

That's literally it. I know lots of more advanced strategies around "getting started is harder than getting going" but they often require multiple steps and a bunch of thinking, which just feels overwhelming and impossible. So I pretend that "twitch my pinky finger" is step 1 on that list, and I just do it, with no particular thought to step two.

It obviously doesn't *always* work, but the hit rate is shockingly high.

When you realize being reluctant to support the war machine is worse than committing war crimes.
@freeschool@qoto.org

RE: https://floss.social/users/mcepl/statuses/112156673028748514

BREAKING: Vatican bans YAML

Oh yeah! I finally got an owl! I've lived here for 8 years, in the middle of nowhere. I hear owls constantly but haven't seen one. The smoking hot wife asks if I want to go for a walk, I say no, then yes.

This bard (?) owl was sitting in the tree watching us. It passed on a rabbit, but watched me, from about 20 feet away.

I feel so full of hope. Everything will be okay. In a world where everything can menace a man, and he cheat death, we really can do anything so long as we are dedicated enough.

Life is a calling to a beautiful enterprise. Find that one which makes your spirit sing and pursue it without relent.

I am not an American, I do not like Trump particularly much for anything other than a performing monkey.

But this one pic of trump; standing strong, absolutely defiant, fist aloft. The secret service clinging to him as if he were a rock in a raging sea. Behind him is the sky; limitless & all ecompassing. The american flag waves above him, almost appearing to float in the air, held aloft by his will, and upraised fist. Spared only by a brief turn of his head.

That is what Divine Mandate looks like. This image is what it would look like to see Aurelian crowned Restitutor Orbis. This is a moment of sublime, transcendent beauty, captured forever in a lens.

What a wonderful, tantalizing image. It reaches into the soul, whispers in its’ ear ‘This could be your nation, too.’

The United States of America is the only extant nation that could ever lay claim to be the Inheritor of Rome. What delightful political drama.

Trump survived assassination attempt. So what? Brandon survived several of them! Remember how them evil people put him on a bicycle? How they made him climb the stairs? Never forget!

Wrote doc/why-did-i-not-think-of-this.txt immediately after returning from the toilet. It's mostly codebase-specific technical notes. Nothing that would make sense to anyone besides me.

I've been unhappy with the amount of overhead the ingestor introduces, right? It's the thing that gets in my head and stays there, it's kinda burrowed in there. (Maybe because it is more fun to do that sort of thing; the words "fucking" and "bullshit" appear way more often in the ActivityPub module than the bits that deal with the network or with the encoder.) Bear with me a moment: blocks aren't just a chunk of storage in this case, blocks are additional network traffic, they're an increase in false positives in the bloom filters, they're filesystem overhead, they're slots in the cache, it's good to minimize them, you will have to take my word for it if it seems not worth considering.

I had this stupid normalization process because I was thinking of things the way you think of them in Postgres, here are the values, we split them out and put an ID in there, etc., and in the mean time had this byte-level encoder, and it picks boundaries around block sizes and then does some of the trivial stuff: truncating around zeroes, headerless deflate. So it is really like how you'd structure it in a conventional application: you pull some object out, say you have `"to":[$long_ass_hellthread_list]` and you turn that into `"toID":"$some_sha256_id"`. Not pretty, and requires that you use accessors to handle things uniformly (i.e., so that the "to"/"toID" split doesn't need to be handled elsewhere in the application), and amounts to a weakref. Weakrefs are fine here, but they're slower: you don't see it until you're parsing the JSON, which means the network trying to fill in pieces when it expects you'll be asking for them or propagating pieces that you use more often, it can't do much with weakrefs, it doesn't parse the JSON.

On the other hand, being smarter about the encoding means that instead of shifting the reference up a level so that we've got to worry about that at the level where we're dealing with AP objects and activities, we push it down one more level. The way the encoding works is that it's split into blocks, then a header is generated for each of the blocks, the headers are concatenated to produce the top of the tree, and this process is recursive. (I don't remember what it actually does, but I recall tarsnap describing some cleverness around finding optimal block boundaries.) It actually doesn't matter to the decoder *where* the boundaries are: instead of storing the string "abcd" as four bytes and a metadata block (48 bytes total), you can store it as four one-byte strings plus four headers, and then another header up one level (4+(4*44)+44=224 bytes), and the decoder handles it fine. That's obviously contrived to be perverse, right, but you could engineer it such that you actually try to find some reasonable boundaries. Take a hellthread for example: the to/cc list is going to be the same, a thousand times (or if you sorted it, it would be). Take a post from Mastodon: there's going to be "content" and "contentMap" and in the overwhelmingly likely case, "contentMap" is going to have only one entry and it's going to be the same as the value for "content". (That's two copies of the Bee Movie.) I was accounting for the to/cc stuff, but there's "content", there's "contentMap" on Mastodong, there's "source" on Pleroma, and that's before investigating the weird ones.

I think it's probably *possible* to cram this into the MarshalJSON() functions, so you can tell the encoder "Hey, the boundaries are here and here", basically compiling a list of hints for it to break on. That would require some sort of side-channeling, it'd require planning and accounting for all of the weird shit all of these servers do (including different versions) and that is a pain in the ass. Just telling the encoder "This is well-formed JSON" and letting it pick things out, that's easier. It can find long strings and arrays, and this ends up transparent to the decoder, but also transparent to everything upstream from the encoder, so the litepub.Activity doesn't need to try to worry about the encoder any more (beyond sorting), I don't need to worry about which things are likely to be redundant, etc. I also don't have to implement it yet; I can delete code that is already written, code that adds unnecessary complications, and then add this stuff later, like if I get bored while watching the import start from the top again. (136,127,584 objects, 161,084,319 activities, 717,471 actors. It does take a minute.)

Obviously, the heuristics will need to be clever, but the worst-case is that they make sub-optimal breaks. It's also the case that we lose single-pass encoding, since it'll have to backtrack, but that's not a huge problem for an encoder and keeping some JSON in memory isn't so bad; if we're normalizing it, then we already parsed it, it's already in memory, so it's not going to be pathological (that is, if it were going to crash us, it would have done so before we parsed it). So that's the downside, I think it's not so bad.

Anyway, if I sound like I've missed something important (I don't think I have, but I'm uneasy), that's there, say some words. In the mean time, I think I can gut a bunch of conditionals before the next rerun of the ingestor.

That brings me to the second half of this post, which is less "someone sanity-check this design" and more "here is a fit and/or start as FSE resurrects itself".

I'll also have (probably after I restart the ingestor again and it runs long enough for me to alter it) a patch for Pleroma's lib/pleroma/object/fetcher.ex (around line 83), but the version running on FSE (and probably get a patch that applies cleanly against a reasonably current version), to use the object proxy feature if you want to test that out. It responds to URLs that it already has, obviously, but will also start responding to URLs that the ingestor has ingested while it ingests them (wherever I was running the ingestor; the test instance at screamshitter.club will have the data, although I am running the ingestor on a different machine where I can blow out the data and then bring it back), which means that Fediverse Jesus will be calling the dead to rise from their graves. (It will also start trying to fetch some objects on demand at some point, but it's going to be very conservative about fetching on demand until I can make sure that I'm not going to DoS fedi. Fetches for AP objects will be signed if I remember to turn that on, and in either case, it will be usable as a media proxy fallback, though that will probably require an extra patch, but at minimum, FSE's media/emoji/etc. will be back when you try to view them from here or from another instance that proxies to a Revolver test instance.)

Anyway, I'm pretty excited; there has been a lot of stuff that is somewhat less visible, like the /inbox fork (for example, you may have seen, in some cases, FSE's IP attempt to double-fetch objects with two different UAs, and that is why), all of which is less visible. This is going to be pretty visible, at least from FSE. There is also a surprise. :revolvertan: (There may actually be two surprises but no one will like the other one because it is going to look cooler than it is and I'll say "no, no, that doesn't work yet, it just looks like it does if you look at it from the right angle".)
Altered_Beast_ResurrectionbyBreakbeat_OC_ReMix.mp3

Find out! 🐑

who the hell are you?
i'm the scatman
yub dub dub dub