Fold a street name to the form the fuzzy branch compares on
Source:R/normalize_gazetteer.R
nar_match_fold.Rd[nar_fold()] settles case and accents, which is enough for an equality join. The fuzzy branch needs more, because two of Quebec's spelling conventions put a correct parse and NAR's own spelling of the same street on opposite sides of the name gate:
* **The hyphen is not a distinguishing character.** NAR writes `du Square-Victoria`, `du Curé-Labelle`, `Alexis-Nihon`; people write the words with spaces, and usually without the leading particule. Whole-word containment is exactly the rule that should catch `VICTORIA` inside `du Square-Victoria` – and it does not, because with the hyphen in place `SQUARE-VICTORIA` is one word. Folding it to a space is what lets the rule fire. This is not a Quebec-only change: `du Bord-du-Lac–Lakeshore` and `Grande Côte` are the same problem, and English Canada's hyphenated names gain the same way. * **`ST` and `STE` are abbreviations of `SAINT` and `SAINTE`**, and NAR spells them out. `ST-JACQUES` against `Saint-Jacques` is six edits on a thirteen-character string – nowhere near the gate, and nowhere near the top of a similarity ranking either. Expanding both sides is the only thing that makes them meet, and applying it to both sides is what keeps it safe: a name that really does contain a bare `ST` still matches itself.
The apostrophe goes the same way as the hyphen, for the same reason: `de l'Orme` and `DE L ORME` are one street.
Applied to the *probe* it produces `match_fold`, which is deliberately a second column rather than a replacement for `name_fold` – the exact branch joins on `name_fold` through an index, and this expression would defeat it.