Deconstructing and rebuilding Wordle (part 2)

In the first installment, we deconstructed Wordle and then built a wordle-tile custom element. We'll now build a wordle-row.

Five in a row

As with a tile, the role of that element will only be about displaying data, with no interaction. The element will therefore be responsible for managing its internal content (shadow tree) depending on its inputs (attributes and properties) to make it respect the game invariants.

A row contains 5 tiles. It can be either empty, as a placeholder, the current row, or an evaluated guess. When in the empty or evaluated state, a row is immutable: either empty or fulfilled. The current state is more complex.

An empty row only contains empty tiles. An evaluated row only contains evaluated tiles (either correct, present or absent). The current row can contain a mix of empty and unevaluated tiles, and a current tile, where evaluated tiles are always at the beginning, empty tiles at the end, and the current tile in between. The current row can be entirely empty in which case the first tile is the current tile, or filled up, in which case all tiles are unevaluated tiles. Letters can thus be added or removed, and this moves the current tile as the row fills up.

g gu gue gues guess

When in the evaluated state, all letters must have an evaluation.

To model those states, we'll use the element's content for the letters, a current boolean attribute to help distinguish an empty row from an empty current row, and a way to set the evaluations.

Those elements can be combined in invalid, conflicting ways, so we need to define ways to resolve those conflicts, with priorities among them, similar to what we did with the wordle-tile element where a state=current is ignored for a non-empty tile, and other state attribute values are ignored for empty tiles.

Let's start with the easy, non-conflicting situations. The default state for an element (with no content and no attribute) will be the empty state. When it has no content but a current attribute, it's in the current state. When its content is longer than the maximum allowed letters, extraneous letters are ignored (similar to the wordle-tile). When its content is long enough, is doesn't have a current attribute, and it has enough evaluations for all letters, it's in the evaluated state.

Now for the invalid situations. What to do if the row is fulfilled, has evaluations for all letters, but also has a current attribute? We'll treat it as the current row and ignore the evaluations. What if there aren't enough evaluations for all letters? We'll ignore the evaluations. But then what to do if there's content but neither evaluations (or too few) nor a current attribute? In this case we'll ignore the content.

To sum up, as soon as there's a current attribute, it's the current row and evaluations if any are ignored. In the absence of a current attribute, either there are enough letters and enough evaluations and then it's an evaluated row, or we ignore everything and treat it as an empty row. Note that things are only ignored, i.e. they don't change the display, but they're still there, so adding a missing item can change its state non-linearly. For instance, if the row has enough content and evaluations but is missing both the current attribute and one evaluation, it'll be in the empty state.


Adding the current attribute will turn it into a fulfilled current row.


If instead one evaluation is added, it becomes an evaluated row.


Finally, to make it more flexible, we'll actually make the number of tiles configurable, through a length attribute and reflected property. We can also make the tile element name configurable, so users can provide their own tile element with a different rendering, provided it respects the same contract as wordle-tile.


In terms of implementation, because we'll have to use the length at different places, we'll change the way we parse its value from the attribute by memoizing it: the property setter still modifies the attribute, but we'll now listen for the attribute change and store its parsed value in a private field, that the property getter will then directly return.


By the way, we'll also handle the shadow tree much differently from the the wordle-tile: because the rendering (updating the shadow tree) can be quite heavyweight, we'll try to do it only once per event loop, rather than each time a property changes. Imagine the situation where the element is dynamically created, so it starts empty, with no evaluations and with length 5. For some reason, the code first sets current to true, then the content to a 6-character long value, and finally the length to 7. We don't want to first change the state of the first tile to current, then undo it and set each tile's letter, and finally add two more tiles, set the sixth one's letter (the one that was previously ignored because it overflowed) and the seventh one's current to true. Instead, we'll batch those changes into a microtask and do the rendering once all properties have been updated. To do that, we'll listen for all the changes (to attributes using attributeChangedCallback, and to the content with a MutationObserver) and queue a microtask when they happen, using a flag to avoid queueing more than one.