invalid geometries from negative buffering in rgeos/sp

Using scot_BNG from the rgdal package:

dsn <- system.file("vectors", package = "rgdal")[1]
scot_BNG <- readOGR(dsn=dsn, layer="scot_BNG")

and buffering with a sufficiently large negative width:

b=gBuffer(scot_BNG,width=-10000,byid=TRUE)

produces an object that gives an error on plot:

plot(b)

Error in checkSlotAssignment(object, name, value) :
 assignment of an object of class ?NULL? is not valid for slot
"coords" in an object of class ?Line?; is(value, "matrix") is not TRUE

Which I think is because the negative buffer has shrunk a feature to nothing.

b[1,] is fine but:

b[2,]

Error in SpatialPolygons(x at polygons[i], proj4string = CRS(proj4string(x))) :
 INTEGER() can only be applied to a 'integer', not a 'NULL'

I'm not sure what the right thing to do here is. Either return a valid
zero-area polygon (if that's not a contradiction in terms) or drop the
degenerate cases?

Barry

On Thu, 3 Oct 2013, Barry Rowlingson wrote:

Using scot_BNG from the rgdal package:

dsn <- system.file("vectors", package = "rgdal")[1]
scot_BNG <- readOGR(dsn=dsn, layer="scot_BNG")

and buffering with a sufficiently large negative width:

b=gBuffer(scot_BNG,width=-10000,byid=TRUE)

produces an object that gives an error on plot:

Yes, but ...

The user gets a SpatialPolygonsDataFrame object with the data slot intact,
but some of the polygons are clearly now invalid. Should we do book-keeping
to only retain the data slot rows for valid polygons, only return the subset
of valid polygons, or proceed as now, and have the user do:

plot(b[gArea(b, byid=TRUE) > 0,])

What makes more sense? I'm open to suggestions ...

Roger

plot(b)

Error in checkSlotAssignment(object, name, value) :
 assignment of an object of class ?NULL? is not valid for slot
"coords" in an object of class ?Line?; is(value, "matrix") is not TRUE

Which I think is because the negative buffer has shrunk a feature to
nothing.

b[1,] is fine but:

b[2,]

Error in SpatialPolygons(x at polygons[i], proj4string = CRS(proj4string(x)))
:
 INTEGER() can only be applied to a 'integer', not a 'NULL'

I'm not sure what the right thing to do here is. Either return a valid
zero-area polygon (if that's not a contradiction in terms) or drop the
degenerate cases?

Barry

--
Roger Bivand
Department of Economics, NHH Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no

A more general fix might be to support NULL geometries in Spatial objects.

Robert

On Thu, Oct 3, 2013 at 6:29 AM, Roger Bivand <Roger.Bivand at nhh.no> wrote:

On Thu, 3 Oct 2013, Barry Rowlingson wrote:

Using scot_BNG from the rgdal package:

dsn <- system.file("vectors", package = "rgdal")[1]
scot_BNG <- readOGR(dsn=dsn, layer="scot_BNG")

and buffering with a sufficiently large negative width:

b=gBuffer(scot_BNG,width=-10000,byid=TRUE)

produces an object that gives an error on plot:

Yes, but ...

The user gets a SpatialPolygonsDataFrame object with the data slot intact,
but some of the polygons are clearly now invalid. Should we do book-keeping
to only retain the data slot rows for valid polygons, only return the subset
of valid polygons, or proceed as now, and have the user do:

plot(b[gArea(b, byid=TRUE) > 0,])

What makes more sense? I'm open to suggestions ...

Roger

plot(b)

Error in checkSlotAssignment(object, name, value) :
 assignment of an object of class ?NULL? is not valid for slot
"coords" in an object of class ?Line?; is(value, "matrix") is not TRUE

Which I think is because the negative buffer has shrunk a feature to
nothing.

b[1,] is fine but:

b[2,]

Error in SpatialPolygons(x at polygons[i], proj4string = CRS(proj4string(x)))
:
 INTEGER() can only be applied to a 'integer', not a 'NULL'

I'm not sure what the right thing to do here is. Either return a valid
zero-area polygon (if that's not a contradiction in terms) or drop the
degenerate cases?

Barry

--
Roger Bivand
Department of Economics, NHH Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no

On Mon, 7 Oct 2013, Robert J. Hijmans wrote:

A more general fix might be to support NULL geometries in Spatial objects.

Maybe, but we do not have in sp a class for mixed types of geometries
either,

so we would have to define NULL of everything. OGC SFS has NULL, but
I would argue that NULL geometries have no analytical meaning, like a point
with no coordinate values - so we don't know where it is. It doesn't fit the
Spatial class, unless we open for a bounding box with NA corners.

In rgeos,
the typical output if there is no positional data is the R NULL value. One
might go towards objects with mixed geometries (GDAL is permitting them for
the drivers that have them in the development version), but I don't see an
analytical need.

How often does, for example, newdata for predicting from a
geostatistical model include both points and lines?

Roger

Robert

On Thu, Oct 3, 2013 at 6:29 AM, Roger Bivand <Roger.Bivand at nhh.no> wrote:

On Thu, 3 Oct 2013, Barry Rowlingson wrote:

Using scot_BNG from the rgdal package:

dsn <- system.file("vectors", package = "rgdal")[1]
scot_BNG <- readOGR(dsn=dsn, layer="scot_BNG")

and buffering with a sufficiently large negative width:

b=gBuffer(scot_BNG,width=-10000,byid=TRUE)

produces an object that gives an error on plot:

Yes, but ...

The user gets a SpatialPolygonsDataFrame object with the data slot
intact,
but some of the polygons are clearly now invalid. Should we do
book-keeping
to only retain the data slot rows for valid polygons, only return the
subset
of valid polygons, or proceed as now, and have the user do:

plot(b[gArea(b, byid=TRUE) > 0,])

What makes more sense? I'm open to suggestions ...

Roger

plot(b)

Error in checkSlotAssignment(object, name, value) :
 assignment of an object of class ?NULL? is not valid for slot
"coords" in an object of class ?Line?; is(value, "matrix") is not TRUE

Which I think is because the negative buffer has shrunk a feature to
nothing.

b[1,] is fine but:

b[2,]

Error in SpatialPolygons(x at polygons[i], proj4string =
CRS(proj4string(x)))
:
 INTEGER() can only be applied to a 'integer', not a 'NULL'

I'm not sure what the right thing to do here is. Either return a valid
zero-area polygon (if that's not a contradiction in terms) or drop the
degenerate cases?

Barry

--
Roger Bivand
Department of Economics, NHH Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no

--
Roger Bivand
Department of Economics, NHH Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no

On Mon, Oct 7, 2013 at 5:51 AM, Roger Bivand <Roger.Bivand at nhh.no>
wrote:

On Mon, 7 Oct 2013, Robert J. Hijmans wrote:

A more general fix might be to support NULL geometries in
Spatial objects.

Maybe, but we do not have in sp a class for mixed types of
geometries either,

True, but that is a different issue.

so we would have to define NULL of everything. OGC SFS has NULL,
but I would argue that NULL geometries have no analytical
meaning, like a point with no coordinate values - so we don't
know where it is. It doesn't fit the Spatial class, unless we
open for a bounding box with NA corners.

There have been cases where I thought it would have been handy,
for example when geometries (e.g. point locations) are known for
some, but not for all records in a data set. However, addition this
type of support would add a significant amount of overhead to all
functions (check for / remove NULL values; NA bounding box vaues).

In rgeos, the typical output if there is no positional data is
the R NULL value. One might go towards objects with mixed
geometries (GDAL is permitting them for the drivers that have
them in the development version), but I don't see an analytical
need.

Me neither

How often does, for example, newdata for predicting from a 
geostatistical model include both points and lines?

Roger

Robert

On Thu, Oct 3, 2013 at 6:29 AM, Roger Bivand
<Roger.Bivand at nhh.no> wrote:

On Thu, 3 Oct 2013, Barry Rowlingson wrote:

Using scot_BNG from the rgdal package:

dsn <- system.file("vectors", package = "rgdal")[1] 
scot_BNG <- readOGR(dsn=dsn, layer="scot_BNG")

and buffering with a sufficiently large negative width:

b=gBuffer(scot_BNG,width=-10000,byid=TRUE)

produces an object that gives an error on plot:

Yes, but ...

The user gets a SpatialPolygonsDataFrame object with the data
slot intact, but some of the polygons are clearly now
invalid. Should we do book-keeping to only retain the data
slot rows for valid polygons, only return the subset of valid
polygons, or proceed as now, and have the user do:

plot(b[gArea(b, byid=TRUE) > 0,])

What makes more sense? I'm open to suggestions ...

Roger

plot(b)

Error in checkSlotAssignment(object, name, value) : 
assignment of an object of class ?NULL? is not valid for
slot "coords" in an object of class ?Line?; is(value,
"matrix") is not TRUE

Which I think is because the negative buffer has shrunk a
feature to nothing.

b[1,] is fine but:

b[2,]

Error in SpatialPolygons(x at polygons[i], proj4string = 
CRS(proj4string(x))) : INTEGER() can only be applied to a
'integer', not a 'NULL'

I'm not sure what the right thing to do here is. Either
return a valid zero-area polygon (if that's not a
contradiction in terms) or drop the degenerate cases?

Barry

-- Roger Bivand Department of Economics, NHH Norwegian School
of Economics, Helleveien 30, N-5045 Bergen, Norway. voice:
+47 55 95 93 55; fax +47 55 95 95 43 e-mail:
Roger.Bivand at nhh.no

-- Roger Bivand Department of Economics, NHH Norwegian School of
Economics, Helleveien 30, N-5045 Bergen, Norway. voice: +47 55 95
93 55; fax +47 55 95 95 43 e-mail: Roger.Bivand at nhh.no