Fix typo breaking st_as_sfc.pq_geometry() on every input

[guslipkin]

Fixes #2625.

Summary

st_as_sfc.pq_geometry() (R/db.R) forwards to st_as_sfc.WKB() using the argument name spatiallite, which doesn't match st_as_sfc.WKB()'s actual spatialite parameter. The unmatched named argument falls through st_as_sfc.WKB()'s own ... and rides along into its final st_sfc(ret, ...) call, where sfc_unique_sfg_dims_and_types() trips over it and fails with a misleading "object(s) should be of class 'sfg'" — unrelated to the actual geometry being converted.

This means st_as_sfc() has been broken for every pq_geometry input — i.e. every PostGIS geometry column read via RPostgres::dbGetQuery(), the exact workflow #1195 introduced this method for — since that method was added. The typo was already present in the fix suggested in that issue and appears to have been merged verbatim.

Why this was dormant until 1.1-3

The typo itself has been there since #1195 (2019), but it was harmless for years: in 1.1-2, st_as_sfc.WKB()'s final calls did not forward ... into st_sfc() at all —

st_sfc(ret, crs = crs)
...
st_sfc(ret) # leave attr srid in place: PostGIS srid that is not an EPSG code

— so any stray/unmatched argument (including the mistyped spatiallite) was silently dropped and never reached st_sfc().

1.1-3 changed this in R/wkb.R to forward ... through:

st_sfc(ret, crs = crs, ...)
...
st_sfc(ret, ...) # leave attr srid in place: PostGIS srid that is not an EPSG code

That's commit a264e6f ("1.1-3 CRAN submission"), landed alongside #2622's precision= round-trip work for st_coordinates() — almost certainly to let callers pass st_sfc()-level args like precision through st_as_sfc(). Reasonable on its own, but it's what first gave the six-year-old typo an actual path to break something: now the stray spatiallite = FALSE rides along in ... straight into st_sfc()'s sfg-class validation.

Confirmed directly against the CRAN source tarballs for both versions (R/wkb.R diff, 1.1-2 → 1.1-3 is exactly the two lines above) and reproduced the crash appearing/disappearing across a clean reinstall of each release.

Fix

One-character rename, spatiallite → spatialite, so the argument actually reaches st_as_sfc.WKB()'s matching formal instead of leaking through ....

Test plan

• Added a DB-free regression test in tests/testthat/test-wkb.R reusing the existing WKB hex fixture from that file, just re-classed as "pq_geometry".
• Verified against the actual compiled package (devtools::load_all() + testthat::test_file()):
  • With the typo (pre-fix): the new test fails with the exact reported error, full backtrace through st_as_sfc.pq_geometry → st_as_sfc.WKB → st_sfc → sfc_unique_sfg_dims_and_types.
  • With the fix: test-wkb.R passes in full.
• Also verified against clean CRAN installs: reproduces on a from-source 1.1-3 install, does not reproduce on a from-source 1.1-2 install — consistent with the R/wkb.R diff above.

Related

• st_as_sfc.pq_geometry #1195 — original feature request; the suggested implementation already had this typo.
• st_as_sfc.pq_geometry() crashes with "object(s) should be of class 'sfg'" due to spatiallite/spatialite typo #2625 — this bug, filed with a minimal DB-free reprex and full root-cause trace.
• a264e6f — the 1.1-3 commit that started forwarding ... into st_sfc(), surfacing the dormant typo.