Where parallels cross

Interesting bits of life

Moldable Emacs: making molds a little easier to write

Too long; didn't read

Molds definition :given and :then don't accept lambdas anymore: you want to declare a function with :fn instead. This small (breaking) change will bring some benefits. If you have already defined your molds, here you will find how to move them to the new format.

The problem

I am pretty proud of moldable-emacs. Lately I am creating molds for making my work simpler. For example, just the other day I shared with my colleagues how I collect production errors with a mold. I pull the errors from our logging system and then I analyse them with another mold. The idea is to bring that report into my editor. Soon (I hope) Emacs will point to me which file I have to fix to get rid of a given error!

By the way, a reminder of the power of moldable development: just showing my little (rough) analysis prompted a colleague to solve an issue with a feature we are releasing! Again a custom view can save the day (and time).

When molds gather data from other systems, they often block Emacs. So, I devised a little pattern to use the great async.el to keep my Emacs usable. Anytime a mold needs slow-to-get data, I start a process that does that in the background. In the meanwhile Emacs fetches the data, my mold displays a message like "Loading some data" in the output buffer.

This pattern works, but it is tiring to code and easy to get wrong. And since the molds I make for work will ping all sort of services for data, I surely need a better solution.

Would it not be amazing to just tag mold sections to say run this async?

It is a problem indeed

This is a more general issue. Each bit of a mold definition could carry metadata useful for running it. My first design of molds is not declarative enough. I decided to use a property list to make sure that molds are data that Emacs can interpret (and run). I now realize I left something out though.

This is how a mold currently looks like:

 :key "Query"
 :given (lambda () 't)
 :then (lambda ()
         (let ((self (ignore-errors
                         (goto-char (point-min))
                         (eval `',(read (current-buffer))))))
               (sexps (call-interactively 'eval-expression))
               (buffer (get-buffer-create "m/tree")))
           (with-current-buffer buffer
             (setq-local self sexps)
             (me/print-to-buffer sexps buffer)

(This mold is a nice utility: you write an Elisp expression that runs in the current buffer and outputs the results in a new buffer.)

You can see that the :given and :then bits are lambdas. That is a blocker. When I get the :given of the "Query" mold, I have to run it. The more I think about it, the more it seems that :given could be other things than a function. For example, it could be a regexp! We could interpret that to activate the mold anytime that regexp is found in the buffer. Or it could be a function with a flat to run it asynchronously!

Do you see? I feel that in my first design I missed just an important little out that could mean a lot later on.

And there is a solution

Time to improve things! Let me show how the new "Query" mold looks like:

 :key "Query"
 :given (:fn 't)
 :then (:fn
        (let ((self (ignore-errors
                        (goto-char (point-min))
                        (eval `',(read (current-buffer))))))
              (sexps (call-interactively 'eval-expression)))
          (with-current-buffer buffername
            (setq-local self sexps)
            (me/print-to-buffer sexps)))))

The lambdas are gone! The :fn stands for lambda here. The me-mold-1 function (which runs molds) will interpret that as something to run.

So we can imagine (didn't finish to implement that yet though) that to make the :then clause run my async pattern, we could define "Query" like this:

 :key "Query"
 :given (:fn 't)
 :then (:fn ... :async t))

It would be clean, no? Also, now that I think of it, I could make so that users can define their own custom interpretation of tags. A good way to separate concerns, I guess (Aspect Oriented Programming, a remake).

Note another thing: you don't need to create anymore the output buffer manually. Experience showed me that is a poor investment of time. You can still define a custom name by defining a :buffername property for the mold.

Another little feature I introduce is a :let keyword. With that you can define bindings available to both :given and :then clauses. This increase code reuse for some molds. Later I want to make sure that these bindings are evaluated only once (_for now is evaluated twice_), so to spare computations.

The steps to refactor molds for the new format are the following.

  1. replace lambda in :given and :then with :fn
  2. use buffername to use the mold's output buffer in the :then clause
  3. put common variables between :given and :then in :let clause
  4. stop returning buffer at the end of the :then clause

If you want to use the new format already, just use me/mold-1 instead of me/mold to run molds AND put the following setting in your init:

(setq me/files-with-molds (list "<moldable-emacs>/molds/contrib-1.el" "<moldable-emacs>/molds/core-1.el"))

Where <moldable-emacs> stands for your path to the mode. This will load and use the refactored molds.

All in all, a small breaking change! I shall post soon about the improvements this allows!


Now a few interesting doors are open for moldable-emacs to evolve. Please let me know if you have any feedback!

Happy molding!