ElixirConf 2017 - Writing an Editor in Elixir - Ian Duggan
- 1. Writing an Editor in Elixir today we’re gonna edit like it’s 1999
- 2. Just how are you planning to bore me?
- 3. Why do this? Why???
- 4. Ok, get on with it then. (I only have to sit here for 45 minutes)
- 5. My story? (the details of my life are quite inconsequential)
- 6. Initial release in 1974
- 7. Computer user since 1979 (core gamer, dude)
- 8. 1. BLOAD SUPERINV 2. <press play on tape> 3. <wait 10 minutes> 4. BRUN SUPERINV (something like that — I was 5 in 1979)
- 9. 5. PLAY SPACE INVADERS!
- 10. Coding since 1982 10 PRINT “IAN ” 20 GOTO 10
- 11. Editing since 1993 (when I met my first love, Emacs)
- 13. So why not write an editor? (‘cause we need more of those…)
- 14. What makes an editor successful?
- 15. Why do some come and go while others persist?
- 16. Why do neckbeards cling to 1970s tech? (or does it cling to them?)
- 17. Why do hipsters insist on editing through 20 layers of abstraction? (css goes well with a latte, and I’ve got 16 cores man)
- 19. ED
- 20. ED
- 21. (still the best on a teletype…)
- 22. vi (also created in 1976…)
- 25. The GUI Era “You’re Welcome” — Java
- 26. Intellij / Eclipse / Netbeans (‘cause I like editing through a peephole)
- 27. "Modern" Editors (yeah, man… air quotes)
- 28. TextMate (because I’m smart enough to know that Notepad is inadequate)
- 29. Sublime (for when you need a break from broken Emacs)
- 30. Atom “I use Atom cause I like mediocre editors” — ElixirConf2017 Trainer
- 31. Please, what’s wrong with my Atom? (the whole point for me is it should just work)
- 32. Some of this may have offended you (and that’s an important clue)
- 33. It’s because, as programmers, we often have a deep, personal relationship with our editor. (we’ve spent more time with it than anyone else!)
- 34. It’s about power and productivity. (sometimes a delicate balance)
- 35. So again, why do this?
- 36. I love Emacs, but things have been rough lately..
- 37. Maybe it’s a syntax thing
- 38. And, as a hacker, I want to write my own.. :)
- 39. And we have this great language that is simultaneously a perfect and a terrible choice for this.
- 40. Great because… High level Easy to develop in Great for orchestration Durability through OTP Code reloading
- 41. Terrible because… Mainly because the GUI ecosystem is pretty limited That just hasn’t been Erlang’s or Elixir’s strength I have arrows in my back to prove this!!!
- 42. But we can fix this!
- 43. Let’s invent on principle. (or rather, in this case, shamelessly steal!)
- 44. A nice correspondence Emacs Emx C core Rust core Elisp Elixir Dynamic typing Dynamic typing Code reloading Code reloading Eval Eval DSLs DSLs Mutable Immutability Beachballs Concurrency
- 46. Basic Design Rust for low-level (where necessary) Back-end (Elixir) Front-ends (text, GUIs)
- 47. Core / Back-end Sessions: have multiple buffers shared or not Buffers: text, files, etc rust-backed for performance (maybe?) Event-maps: map input actions to state transformations
- 49. Sessions
- 50. Front-ends
- 51. Buffers defprotocol for multiple implementations: • Sting-based • Elixir “ropes” • Rust “ropes”
- 52. Buffers
- 53. Buffers
- 54. Buffers
- 55. Buffers
- 56. Event-maps Can be code or data Stateless Allows for easy code- reloading
- 57. Key-code parsing
- 58. Key-code parsing
- 59. Ropes
- 60. Ropes Basically a string as a binary tree Balancable
- 61. Ropes Index(i): character at i, (log N) Concat(S1, S2): concatenate two ropes, O(1) (or O(log N)) Split (i, S): split into two ropes O(log N) Insert(i, S’): insert string at i O(log N) Delete(i, j): delete the substring Ci..Cj, O(log N) Report(i, j): output the string Ci..Cj, O(j + log N)
- 62. Elixir Ropes I updated a library Thank you @copenhas on github. I did a PR. Can I publish it to hex.pm?
- 63. Rust Ropes In development. Makes use of Ropes from the Xi project. Will integrate as Rustler resource object.
- 64. Front-end terminal (tty_sl, others) Wx (:wx provided by OTP) Electron (just kidding, but why not, help me add it?) Exui (maybe? Boyd and I are already talking)
- 65. BEAM/OTP is 500k lines of C code and 2.4 M lines of Erlang code! Atom / Electron is scary?
- 66. Front-end protocol Connect to backend, receive a session Send input events (keys, mouse, etc.) Receive output events, render (sound familiar anyone?)
- 67. escript mix.exs
- 68. mix.exs escript
- 70. erl start options —no-input (don’t read input) —no-shell (don’t start the erl shell)
- 72. escript
- 73. IO.Ansi IO.Ansi didn’t have cursor movement stuff
- 74. Emx.Ansi Maybe it should be core?
- 75. But there were some issues You can't have erlang shell or IEx with it (terminal ownership conflict) I couldn't get IEx —remsh working wth it (even with elixir, mix, iex, etc running)… (someone please help?)
- 76. Maybe a port?
- 77. Do what IEx does? IEx replaces the erlang shell erl … -user Elixir.IEx.CLI
- 78. The erlang shell There’s a great explanation of how it works here: https://ferd.ca/repl-a-bit-more-and-less-than-that.html I’ll summarize it quickly
- 79. “Old” shell
- 80. “New” shell
- 81. IO Protocol Phoenix uses this during code reload. Basically send/receive bytes/ characters as messages.
- 83. Replace tty_sl? Let erlang think it’s in control of the terminal. We’ll secretly control the terminal.
- 84. Run our back-end as a port-driver? It can own the terminal and provide the shell. This is really a variation of the other ideas.
- 85. Use a pty? We’ll can start erlang with a pty so it thinks it owns the terminal. This has the benefit of letting us do other-language front-ends that run a backend over the pty.
- 87. We can fake it with a pty
- 88. We can fake it with a pty
- 89. Rust-based PTY Front-end TODO: show termion code.
- 90. But why another node?
- 91. My old friend, telnet The name stands for "teletype network” Basically “terminal over network” This is a prototype, so start simply. Eventually this should be SSH.
- 92. My old friend, telnet
- 96. Let’s try connecting telnet <my_ip_here> 5555
- 97. Wx If you want to do cross-platform GUI work, your options are really limited They fall into a few basic categories: Looks native on the platform (lowest common denominator, WX) Looks the same everywhere (consistency of implementation over consistency of look)
- 100. :wx.wx_object
- 101. How to wrap the constants
- 102. But this is tedious
- 103. So I wrote a library.
- 104. And then I found this post ! :)
- 105. WxWidgets code wrapping helpers for records and constants adapt for elixir idioms
- 106. Very basic, still needs work :wx wraps WxWindows directly… so… All the arguments are in the wrong order! Makes fluid interfaces hard
- 107. Also needs docs I think I will find some way to wrap and import the docs from WxWindows and attach them to the right methods
- 108. Code wrapping
- 109. Code wrapping Tell me if there’s a better way, but I think macros might obscure too much…
- 110. use WxObject
- 111. And voila!
- 112. wx_const(:FOO)
- 113. WxObjectServer
- 114. Supervising (YMMV, not sure how well this works yet…)
- 115. Let’s try it out.
- 116. TIL (please learn from my mistakes)
- 117. Don’t spend days fighting with regexs, use a parser. (you are clearly using the wrong tool)
- 118. Combine is great, but it still needs laziness.
- 119. Lazy Parsing in Combine
- 120. Do not name your top-level module Wx (you’ll spend hours wondering why things don’t work — I still don’t know why)