detecting browsers is annoying

i discovered the issue when dotsong was trying to scrobble “Instagram”. i had to design this weird detection logic that runs if the process name matches a list of known browsers:

1. is allow_browsers even enabled in settings?
2. could we find it on deezer

this feels a bit shoddy because for more niche tracks (which is totally possible if you’re playing them from a browser) it probably won’t be found.

this makes me think i should lookup on musicbrainz instead. i also don’t want to rely on a private API, metabrainz seems more trustworthy especially considering spotify’s latest API changes

also: i think i want to add a few custom discord apps for certain common players, like apple music and spotify so it can be “Listening to Apple Music” and stuff yk?

started rewriting the entire pipeline. main changes:

1. complete MediaCenter::process_event rewrite
   rewritten completely from scratch. now takes up 40 less lines, skips invalid tracks earlier, and is much simpler to understand

2. ticking logic improvement
   the app used to tick every 5 s to check track duration, which means 12/min. instead, i now calculate how far the 50% mark is and sleep for only that long; when we’re past it, it’ll just idle until the next track. the app now only wakes once per event, rather than 12 times/min

3. use unicase crate to avoid unnecessary allocationsmost of the String allocations i make in the program are to_lowercase() calls for comparisons. the unicase crate allows me to make case-insensitive comparisons without extra allocations.

all of this SHOULD mean less CPU and memory usage (i’m a bit obsessed with that, if you can’t tell yet), but i havent benched anything yet

this PR edits like 500 lines but there’s only like 20 lines more than there were originally lol

i had been initializing a whole new last.fm/libre.fm client on every now playing request/scrobble, which was unnecessary and clunky. so i now:

1. store a static CLIENT_POOL: LazyLock<DashMap<String, Client>> where the keys are the Scrobbler ids
2. get the client for this scrobbler or create a new one and store it. this only runs on new requests, so we don’t unnecessarily create client.
3. remove the client when the scrobbler is dropped, e.g. when the config is updated

i’ve never used drop logic before but i think its like the coolest thing ever, so i was excuted about that

originally, i had written two completely separate track detection pipelines for macOS and windows/linux, but i realized that it resulted in a lot of redundant logic, and making changes to that logic required testing on both macOS and windows for things that shouldnt have (e.g. cover art uploading), so i decided to unify it into one processing pipeline, and the platform-specific parts are just getting it to be a MediaInfo type as soon as possible.

by the end of it, there were only 20 more lines than there were before, but there is now only one processing pipeline which is a huge help for implementing stuff in the future!

also, some more tangible things:

• dark mode UI
• dotsong will do its best to filter out tracks from browsers instead of just taking it with no questions
• arm64 builds for linux now!

finally merged the cover art uploading feature! works on all platforms (linux is untested, but… we’ll get there)

fixed a really obnoxious bug i’ve been dealing with pretty much the whole time. on macOS apple music, when you pick a new track, it’ll broadcast a pause event and THEN a play event immediately after.

for whatever reason, pause events always LOSE the race against the play event, and therefore overwrite the play event. it probably has to do with deezer caching.

the fix was just short circuiting; check early if this is a pause event and we alr have the info; if so, dont worry about processing anything else

published another two versions today!

first update contained a lot of those optimization fixes from before, but the second one was me testing with a user on windows (i normally use macOS, so testing on windows isn’t the easiest)

now i’m doing testing on debian which is not working in any way at all </3

created a litterbox mock api with bun so i can test against image uploading against it

compliant-ish. enough that it only accepts valid parameters and returns realistic URL responses.

and its only 57 lines! i <3 bun forever

the source is here https://gist.github.com/thrzl/22cde52746d4febf2c513b7fc7b829eb

more subtle improvements

in this session, i removed some unnecessary ui elements (like a redundant separator and useless buttons)

the more interesting thing is i fixed a bug where i was loading stale states, which cause it to spam

now, i’m about to test AppImage builds, whenever i feel comfortable enough to put out another release

spent a LOT longer than 20 minutes trying to debug this weird memory leak that happens whenever i even ACCEPT image data on track events

i literally gain a whole 200 MB PER EVENT, an “event” being any time the track info doesn’t stay the same. so if you pause it, play it, or the track changes for any reason, that’s an event, and it stacks

so its totally unreasonable rn for me to get the images cus bro?? and i know its not in my code because i didn’t even add the handling yet i literally just accept the image data and it does that

so i just filed an issue in the upstream crate pray for me guys

today i was trying to implement scrobble retries but i ended up not liking the current shape. i was gonna write to a file and send scrobbles later next time the host succeeds, but that was super annoying to implement

i think i might end up trying a threads based approach where i just spawn a new thread whenever a scrobble fails; but that wouldnt handle network conditions as well.

i think ima take a break from that and look at getting image support working tomorrow. that needs me to fix the upstream media-remote crate to fix a memory leak, first and foremost; then i’d want to lazy load images cus 70 MB just for getting a track is absurd

lots of code cleanup

learned today that you can’t just use Arc<Mutex<...>> everywhere (well you can but its not necessarily a good thing) so i’m now using RwLock and ArcSwap which makes things a lot easier. i ended up changing tons of internal stuff in basically every file and still my PR only added like 30 something lines