Discussion
Papyrix, being some ereader (specifically Xteink X3/X4) firmware, must fit the paragraphs you wish to read onscreen! So today I'll continue my study of its text-wrapping code!
Naturally I'll be wanting to port this code for my "merephone" project, so your SMS texts fit onscreen.
A faster codepath that can scale to excessively large paragraphs iterates over batches of parsed words, with an abort-callback checked in each iteration.
1/?
For each word this fancier text-wrapping algorithm periodically checks an abort-callback, checks if recorded demerits exceeded a limit, iterates over subsequent words until the line has filled using helpers to scores recording the minimum penalties in a Dynamic-Programming array.
Alternatively a greedy algorithm validates that there's words to layout before iterating over each one. For each word it periodically checks the abort-callback & gathers words until we've filled a line.
3/5?
Once we've filled a line it checks if the next one would overflow too in which case it considers (via a helper) hyphenating the word & keeping part of it on this line. Regardless it records the line break.
In either of these 2 codepaths it appears the performance is more about RAM-space than CPU-time, they're both reasonably fast!
That hyphenation helper defers to a helper library for choosing hyphenation points & validates if there are any, followed a reimplemented greedy line layout.
4/5!
Then there's logic for converting the results into the render tree. For each line a helper retrieves the line breaks, sums onscreen width, possibly computes a wide a space should be to justify the text, finalizes chosen alignment, branches over text-direction to finalize positions & (via a helper) hyphenations of all words, removes processed data from arrays, & defers to a callback.
4.5/4.5 Fin for today! Tomorrow: "block" layout objects!
This does look like great code I'll want to reuse!
Until no words are left in this fast-path it grabs a maximum number of words to wrap in a recursive layout-object & recurses into these layout-methods, reclaims any words not laid out, & propagates completions or errors.
The core layout logic may involve validating that there's words to layout, constructing some lookuptables, iterating over all words, reformatting results into chosen line breaks & reversing that array, & falls back to single-word per line.
2/?