tickling bash, the quest for the perfect menu

2020-06-19 • 2020-05-10


(where the 'perfect menu' means you can start typing right away)

tl;dr: I wanted to stop reading stdin after the first newline to stdout, and I did a hacky thing to get that, the essence of which is:

sh -c 'echo $$; echo a; sleep 100; echo b' | (read pid; read line; echo "$line"; kill $pid)

I really like selecting things from fuzzy searchin' keyboard-driven menus. There's something magical about being able to bring up a menu, mash a few buttons, and get to where you are going. There are lots of really cool generic "select thing from list" programs out there. As time goes on, you think of more things you might do, and some take it really far.

1 Setup

Let's go through some history first. dmenu2 was my goto 'selector' for years, because it has some neat features like positional arguments and the ability to specify the colorscheme from Xresources. In my day to day, occasionally I would find a neat use or have a cool idea, and then integrate it into my setup. Stuff like "search mpd songs and insert one at the current playlist position" or "select a password to copy into the clipboard". Wow this is all super cool! We can just do things and integrate them with little script ideas, neat.

Eventually I had a hankering for a "mega dmenu script". I'd seen rofi here and there, and one of it's features was a builting "switch to window by title" feature, which is pretty cool! I right away stole the idea in a dmenu script. But then I got to thinking, what if I had a "switch to everything" script. The vision I had in my head of a single hub to jump to or DO anything was a really nice picture, indeed. And thus, dmenu_switcher was born, a script I would then shout about many times. dmenu_switcher is a script that matches dmenu selections to bash functions in an associated array – so I added actions to:

  • jump to window
  • jump to browser tab
  • jump to open file in emacs

In the process of implementing that last point, I realized I might not need dmenu after all. Within emacs itself, I was using ivy to switch to things. From there I learned that you can spawn an emacs frame that is solely a minibuffer, and that meant the next step was to create a dmenu-like script that just spawned an emacs frame containing ivy. In combination with prescient.el this means my "mega menu" gets extra sorting based on selected entries – previously selected things automatically float to the top! How nice. However, I'm already using dmenu in many places. This leads to me making a script in my path also named dmenu pointing at emacs_dmenu.

Such is the state of things for a few months.

🍄

In the past few weeks, I've been practicing my touch typing. As you get a little faster, latency starts to matter. If emacs is lagging it's like I'm typing through the mud (tangent: emacs was lagging and I found out why). You start noticing things. Things like the fact that dmenu (and my emacs_dmenu wrapper) BLOCK UNTIL STDIN IS READ. After performing some tweaks I was down to ~100ms for gathering all the jump candidates. emacs takes about another 100ms to create a frame on my machine. A 200ms bump is quite noticable in the "flow" of things (you have an action in mind, you mash the keybind and start typing then OOPS I just sent gibberish to my text editor).

2 Rofi

At this point I decided to check out rofi – and was rewarded with the very nice -async-pre-read flag. Setting the value to 0 means that rofi will start accepting keyboard input right away. That means I can just summon the jump menu and start typing before candidates are ready. This is a golden experience! It means I can include search candidates that take a long time to get (say, my browser history in the past week) but still start typing right away.

3 Tickling bash (cheating)

So, I'm using rofi in dmenu mode, because I'm using it as a dmenu replacement. This means I'm at the mercy of the shell (pipelines waiting for stdin to be read before returning). An example is shown below (rofi pops up, you select the 'a' option before 'b' is ready, you are stuck):

(echo a; sleep 10; echo b) | rofi -dmenu -async-pre-read 0

You see the issue with this? After I select something (intending to move on to an action), I'm stuck again! until all the candidates have been created. Yet another bump in what I'm trying to do. Clearly, this is an insult and I must make the machine do my bidding. This is not a fault with rofi – it returns right away after selecting an option, as expected.

So the way I cheated was to run the 'candidate maker' in a subshell, pass it's pid along to my wrapper, and kill it once rofi returned. Can't pass options when your dead. A shortened version of this fix may be found at the top of this post, but here's some links showing the use in my wrapper (gross):


IndexRootSourceSitemap