Version 1.2-7 (development)

#103 refactoring local joincount test by Josiah Parry
add hotspot methods for localmoran (analytical, permutation, Saddlepoint and exact), localC (univariate and multivariate) and localG (analytical and permutation)
#95 add "two.sided" to lee.mc(), same for sp.mantel.mc(), EBImoran.mc(), joincount.mc(), geary.mc()
#92, #93, #94, #96, #97 contributions of prototype bivariate Moran, local bivariate Moran and local joincount and bivariate joincount tests by Josiah Parry
#91 tolerance.nb() update by F. Guillaume Blanchet
updating coercion for Matrix 1.4-2
fix ncpus issue in dontrun examples
remove suggested packages rgdal, rgeos, maptools

Version 1.2-5 (2022-08-11)

permit use of data.frame or tibble as matrix for functions creating neighbour objects from 2D points (preferred use an object inheriting from `"SpatialPoints" or "sfc")
fix #87 wrong logic in infinite weights in nb2listwdist()
https://github.com/r-spatial/s2/pull/174 speeds up dnearneigh() for geographical coordinates without using s2::s2_closest_edges().
Adapting vignettes for absence of rgdal and maptools

Replace rainbow() by hcl.colors(..., "Set 2") in plot.skater().
Add link to R-sig-geo thread on EBlocal() NaN estimates when many counts are zero on help page.
Revise and add documentation for object returned by localC_perm() #68 #72 #73 #74 #75 #76
localmoran.sad(), localmoran.exact() and localmoran.exact.alt() will now use multiple compute nodes if needed; if Omega is used, multiple cores may need more memory #77
For s2 > 1.0.7, use indexed distances in dnearneigh() https://github.com/r-spatial/s2/pull/162.

Recent changes in poly2nb() had reduced and most recently (1.1-8) removed the use of snap= in finding candidate neighbours; many thanks to Matilda Brown for a clear and well-documented issue #65
Add local Geary’s C #66 thanks to Josiah Parry, discussion on further work on #68
localmoran_perm() returns both look-up and folded rank p-values

In poly2nb(), reverted removal of legacy interpreted overlapping envelope code for sp objects that cannot be coerced to sf without rgeos.
Add Fortran character handling USE_FC_LEN_T WRE §6.6.1.
Checks OK with forthcoming deldir 1.0-0.
Fixes #62 clarifying dnearneigh() help page

knearneigh() and nbdists(); added prototype adaptation to s2 for unprojected coordinates, used if sf_use_s2() is TRUE which became the default for sf 1.0.0 https://github.com/r-spatial/s2/issues/125. These are activated by default.
dnearneigh() can choose the prototype s2 approach if sf_use_s2() is TRUE and use_s2=TRUE for unprojected coordinates; from https://github.com/r-spatial/s2/issues/125 it seems that distance thresholds at present use brute-force rather than spatial indexing. Use is not activated by default.
poly2nb() now uses sf::st_intersects() to find candidate neighbours unless findInBounds= is not NULL. With spatial indexing, this is very fast and scales well for large data sets. If sf_use_s2() is TRUE, sf::st_intersects() passes the geometries to s2::s2_intersects_matrix(), which also uses spatial indexing and is very fast, scaling well for large data sets.
localmoran() and localmoran_perm() return cluster quadrants in an attribute for three splits, zeros, means and medians on the variable of interest and its spatial lag.
localmoran_perm() returns the skewness and kurtosis of the permutation samples.

#55 related to #20 and cycling order in setting up grids provoked re-design of interface to cell2nb(), with passing of "GridTopology" or "SpatialGrid" objects as unnamed first or x= argument. Coerce "RasterLayer" or similar raster, terra or stars objects to sp class objects first if need be.
In working with renewing the arguments to cell2nb(), it was useful to add tinytest support, which is now present for this function and may be extended to other functions for creating "nb" objects.
#58 contributed by Jeff Sauer and Levi Wolf (from https://doi.org/10.31219/osf.io/ugkhp) providing conditional standard deviates for local Moran’s I
Error in assignment to matrix detected by CRAN check in SIDS vignette, section on median polish

Changes to continuous integration and vignettes.
Error in poly2nb(, queen=FALSE) in sf grids (double counting of closed polygon start/end points), https://github.com/r-spatial/spdep/issues/50, thanks to Christopher Kenny.
Adding local Moran and local G conditional permutation: localmoran_perm() and localG_perm().
Adding nb2listwdist() contributed by René Westerholt.
Adding use of sf through GEOS to find polygon contiguity candidates in poly2nb() if geometry count >= 500 - uses intersections in polygon envelopes.
#38, #53 removing RANN, adding dbscan suggestions for fast dnearneigh() and knearneigh() via use_kd_tree= argument for fast planar neighbour set finding in 2D and 3D. Affects soi.graph() too, which had used RANN.
#54 avoid partial matching in glist= handling.
Disambiguating spdep and spatialreg model output object class names prior to making spdep model fitting functions defunct.

Replacing broken geoda URLs, moving knitr to rmarkdown, work-around missing weights files in spData.

A small maintenance update to accommodate a forthcoming change in spData (a dataset used in an example in spdep from spData is changing its name; the name had involved putting “x”, “y” and “xyz” in the global environment through lazy loading a dataset).

Follow-up version of spdep with all the functions and methods transferred to the spatialreg package marked as deprecated, but still exported from spdep. Reverse dependencies passing with the released version still pass for me with this version.