Loads the **Repertoire quebecois des adresses** into the cached NAR database as its own tables, `RqaAddresses` and `RqaStreets`. Nothing in `Addresses` is touched. Once imported, `"rqa"` becomes available as a [geocode()] tier.
Arguments
- version
NAR version whose database receives the tables, passed to [nar_connection()]. Default `"latest"`.
- refresh
Logical. Re-import even when the tables are already present.
- csv
Path to an already-extracted `RQA.csv`. Defaults to `getOption("rqa_csv")`, and downloading the release when that is unset.
Why a separate table
RQA holds roughly 308,000 civic addresses NAR does not, about 9 [normalize_address()] – which has no online fallback – that coverage is the single largest block of what still fails in Quebec. It is nonetheless kept separate rather than merged, for three reasons recorded in `system.file("notes", "quebec-addresses.md", package = "cangeocode")`:
* Merging destroys the only instrument Quebec has. Everything known about what NAR is missing in Quebec is known because the two registers can be read against each other; a merged table can no longer be asked the question. * The added rows are positionally *weaker* than what NAR carries – 20.3 building-placed against 26.9 `Incertaine` by RQA itself – so merging would quietly degrade what `geom_source = 'building'` means for Quebec, which is already misleading there. * A merged table stops being NAR. [nar_provinces()], the row counts in the vignettes and `nar_schema_version()` all describe a Statistics Canada release.
What is imported
Certified rows only (`etat = 'Certifiee'`), which drops about 7,500 retired addresses. The whole register is loaded, not just the rows NAR lacks: the gap is a property of the pair and has to be recomputable, and a table subset against one NAR release would be silently wrong against the next. `IN_NAR` records which side of the gap each row fell on **for the release it was imported into**, since the tables live inside that release's database file.
Street names are reshaped to NAR's convention on the way in, because that is what makes the two joinable: NAR keeps the leading particule inside the street name (`de la Cote-de-Liesse`) and RQA keeps it in a column of its own, so `STREET_NAME` here is particule plus specifique, and the generique becomes `STREET_TYPE`, canonicalized through the same French lexicon [normalize_address()] uses. Comparing the raw columns instead reads 1,265,940 missing addresses where there are 357,723.
Examples
if (FALSE) { # \dontrun{
rqa_import()
con <- nar_connection()
geocode("431 Courtemanche, Montreal-Est QC",
method = c("nar", "rqa", "nar_interpolate"), con = con)
} # }