Even after using a relatively friendly shell 1, I find writing shell scripts a major pain point in automating things. Yesterday, I wanted to make my default browser (whatever that means) open either firefox or chromium based on a regex pattern. What do I need to do this?
- A store for regex patterns mapping to browser executables (in a yaml/json maybe), and
- a way to match and call corresponding executable.
My first thought was to understand if it is possible to do that in xdg-utils
somehow. Then I started to look for a command line yaml parser. Gave up pretty
quickly on that. Not because there are no tools/tricks, but because most of the
them are not very consistently interoperable. One reason might be that these all
have relatively low bandwidth connection (just piped text) in a shell like
environment and that makes the actual programming unnecessarily slow and
attention demanding.
Anyway, a good way to write shell scripts is to just not write shell script but use a regular language. Python is really nice since it's dynamic and has lot of libraries but takes warm up time which is intolerable for tiny scripts. I finally am more or less moving towards Common Lisp for this, which is pretty snappy when using the dumped image. There is this issue with large images (sbcl's starts at around 60 MBs), but I am using a similar approach to fare and will be sharing the same image for all the commands.
The system is here. There is one roswell binary called pod which dispatches
functions like pod <sub-command> [<args>]. While setting up, these subcommands
are written out as shell scripts. I tried using lake for a few setting-up things
and found it to be very neat. In fact any make-ish system which can use your
program's code is going to be nice.
Other than having a language with nice interactive capabilities at my disposal, another specific benefit with Common Lisp comes with the fact that my window manager can import the system and can share a lot of these tools/scripts/functions.
Footnotes:
fish shell