Part of the work of the Friendly Error Fellowship involves resurrecting the original work on p5’s Friendly Debugger, which was disabled due to various limitations documented in #971.
One of the core features of the legacy Friendly Debugger was an argument validator, which would:
- Check that a function call contains the correct number of parameters, and
- Check that a function call contains the correct type of parameters.
This would be useful for beginners in particular. For instance, imagine
passing a number into a p5 function that expects a string; if the function’s
toUpperCase() on the argument, this will result in
TypeError: toUpperCase is not a function being raised, which will be quite
confusing to newcomers. Much more helpful will be to log a friendly error
suggesting the user pass in a string instead.
Now, let’s take a quick detour to look at how p5’s reference documentation is defined. We use a tool called YUIDoc that generates it from some specially-formatted comment blocks in p5’s source code that look like this:
The above is a snippet taken from the source code for
is used to generate its documentation.
As Jason Sigal brought up in #759, it should theoretically be possible to take the metadata extracted from such comment blocks, and leverage it to do some of our type-checking for us, rather than violating DRY, potentially introducing more errors, and complicating our codebase by manually writing extra type validation code.
However, we’re also in a bit of a bind because we don’t actually know if our reference documentation is correct. It would be really unfortunate for the friendly error system to give the wrong kind of advice to beginners, increasing their frustration even more!
So how might we “unit test” our own reference documentation?
Here’s one approach: write a type-checker that uses p5’s own YUIDoc metadata, and use it to type check p5’s own example snippets.
Doing this has a few advantages:
It ensures that our documentation is accurate. For example, if a working example snippet is passing a
p5.Colorinstance as a first argument to
lerpColor()but this isn’t documented as a valid argument type, our type checker will complain. We fix the bug by fixing our documentation.
Conversely, it helps ensure the quality of our example snippets. If our example snippet is calling a p5 API the wrong way, it’s probably broken!
It’s possible that we can reuse the type-checking logic to resurrect the Friendly Debugger’s type-checking system going forward.
I’m now working on one approach to this kind of solution in #1287.