Busqer

Busqer: Simple Spotify client via DBus/MPRIS interface

Developing Notes

• Using Lwt (cooperative thread-based concurrency, stands for lightweight threads), GTK3 (Desktop GUI), and React (Functional Reactive Programming, Lwt_react is lwt-friendly bindings)
• Brief overview
  • Write/Read Spotify’s media player info through DBus/MPRIS
    • e.g. Song Length, Track Title, Album Art Url, If current track changes
    • OBus library exposes DBus properties as React FRP signals.
  • Display this info in a simple GUI
    • Leverage FRP signals: Trigger GTK calls if underlying FRP property changes.
    • Extract colors from Album Art to color GUI
  • Lwt and GTK both require their own infinite main loop, combined via lwt_glib library.
    • Also, React usually needs a manually created loop
• Be careful about mixing Lwt with normal blocking IO (src: https://discuss.ocaml.org/t/concurrent-tasks-in-lwt/14229/5)

(* This immediately returns *)
let () = Lwt_main.run @@
  let open Lwt.Syntax in
  let p1 = (let* () = Lwt_unix.sleep 4. in print_endline "waited 4"; Lwt.return ()) in
  let p2 = (let* () = Lwt_unix.sleep 2. in print_endline "waited 2"; Lwt.return ()) in
  Lwt.return ();;
(* Changing last line to `Lwt.join [p1;p2];;` properly blocks *)

• MPRIS spec: https://specifications.freedesktop.org/mpris-spec/latest/Player_Interface.html
  • DBus is based on Methods, Signals, and Errors. Basically an OOP language.
• Testing with OUnit2 and QCheck2
  • https://cs3110.github.io/textbook/chapters/data/ounit.html
  • QCheck2 has integrated shrinking compared to version 1
  • Dune has features for testing, like Diffing the Result
• React has considerations for concurrent updates
  • To concurrently update values p1 and p2, their respective sets of “descendents” must be disjoint from each other
    • “descendents” are signals and events that are created from p
    • I believe this is because the update step of a value can only handle 1 input
• GTK handles images through GdkPixBufs
  • PixBufs are raw representations, i.e. each pixel is three 8-bit values and a row is N pixels
  • rowstride refers to the byte-width of a row, e.g. N channels * row_pixel_width * byte_size
• OCaml has Bigarrays that are generally used to interface with C & Fortran arrays
  • They don’t have any ways to map, only get/set
• GTK does styling and theming with CSS
  • In 3.0, it uses CSSProvider and StyleContext
  • Usually a widget has CSS classes/ids added/removed to correspond to different UI states and Theme states
  • There is no easy way to change the specific color inside a CSS class at runtime
• Lablgtk are the bindings I use for GTK3
  • the Rocq IDE also uses it (sourceview is for complex multi-line text such as a text editor)
  • However, it was some rough edges Unison proposes to move away from lablgtk
    • I have had issues with GdkPixbuf.get_pixels seg-faulting for no apparent reason, even before manipulating the pointer
• To use Lwt with blocking IO:
  • Wrap in Lwt_preemptive.detach, “It is mostly useful when making calls to a third-party library that does not provide Lwt-aware system primitives.” https://raphael-proust.github.io/code/lwt-part-2.html
• I profiled using perf flamegraphs and the Firefox Profiler viewer
  • A lot of time was spent on Bihist.make_freqs, which was a naive immutable IntMap fold
  • I converted this to a mutable Hashtbl fold with a parallelized Lwt list iteration with some Seq
  • runtime was reduced from ~15s to ~6.5s (Hashtbl) to ~40ms (Hashtbl & Lwt)
  • commit: 2f9fdfaadb8b12067224fd14f55cb93b3e6b84a5
  • perf record --call-graph=dwarf -- ./_build/default/bin/main.exe - standard perf.data file
  • perf script -F +pid > test.perf - convert to file that Firefox Profiler can read
    • https://perfwiki.github.io/main/tutorial/#firefox-profiler

Random Notes

• OCaml’s effects are untyped, i.e. “the compiler does not statically ensure that all the effects performed by the program are handled”
  • As of 250514: Future of OCaml states that Typed Algebraic Effects are a long-term plan
  • Blog post on effects: https://lewinb.net/posts/17_playing_with_ocaml_effects/
  • It seems that EIO is the de facto for effects-based IO
• OCaml has polymorphic variants
  • Discuss OCaml: Is there any kind of guidline about when to use polymorphic variants?
  • Cornell CS3110 notes
  • SO: Variants or Polymorphic variants?
  • SO: Polymorphic Variants cannot use in-line recrods
• OCaml has StdLabels which are versions of some modules where functions liberally use labeled arguments
• Jane Street has a library Incremental
  • They say it is self-adjusting computations, i.e., computations that can be updated efficiently when their inputs change..
  • FRP and SAC are subtely different
  • Umut A. Acar is the researcher behind these ideas
• Reddit: OCaml Features + First-class modules vs Functors
• Discuss Ocaml: How to organize functor-heavy code?
  • User ivg gives good comments like using existentials (simple GADTs) type gps = Gps : 'gps service -> gps
  • Existentials are variables in GADT constructors that are in arguments but not the final return type