Skip to content

Error reading spatialite table with sf::st_read

7 messages · Barry Rowlingson, DUTRIEUX Loic, Edzer Pebesma +1 more

Original
DUTRIEUX Loic
# 
Hi everyone,

When trying to read data contained in a sqlite table with the full spatialite options (see what FORMAT=SPATIALITE implies here --> https://gdal.org/drivers/vector/sqlite.html), I get a 
wkbType: 30078980 Error in inherits(sfc, "try-error") : object 'sfc' not found.

See the example below:

# In bash
ogr2ogr -F SQLITE -dsco SPATIALITE=YES /tmp/olinda1.sqlite ~/R/x86_64-pc-linux-gnu-library/4.0/sf/shape/olinda1.shp

# In R
library(sf)
library(DBI)

con <- dbConnect(RSQLite::SQLite(), '/tmp/olinda1.sqlite')
st_read(con, query='select * from olinda1;')


If I create the database without the SPATIALITE=YES creation option (geometries encoded as WKB), reading with sf partly works (no CRS) but then I lose the spatialite functionalities. Any suggestion?

Kind regards,
Lo?c
Barry Rowlingson
# 
Looks like that error might be generated about here:
https://github.com/r-spatial/sf/blob/5dde39c8261dc6b202e6cde9d41ad3bf0f46aa3a/R/db.R#L134

where the code does:

try(sfc <- st_as_sfc(tbl[[i]]), silent = TRUE)
if (!inherits(sfc, "try-error")) {
tbl[[i]] = sfc
success = TRUE
}

If the first "try" fails then `sfc` doesn't get created and so the error
happens on `if(!inherits(sfc,....)`.

The code is looping over columns trying to find a valid geometry column,
but not trapping this error properly. A bug, I think. Feel free to report
as an issue.

However I don't think that gets us out of the woods - that code only talks
about PostGIS databases, but it gets called for any object of class
DBIObject, which your SQLite connection object is:

 > inherits(con,"DBIObject")
 [1] TRUE

Perhaps that method should be something like st_read.PGConnection if it
only works on PostGIS databases? Experimenting with bits of that function
don't get me the geometry from my test spatialite DB anyway, and I think
its due to it being PostGIS-specific. I kept getting empty geometries.

But why not read the Spatialite directly without going through DBI?  Here's
a test I made of some points:
Reading layer `pts' from data source `/tmp/olinda1.sqlite' using driver
`SQLite'
Simple feature collection with 1000 features and 1 field
Geometry type: POINT
Dimension:     XY
Bounding box:  xmin: 234250 ymin: -198750 xmax: 909750 ymax: 502750
CRS:           unknown

THe CRS is printed as unknown but the WKT is there. Maybe this is a funny
CRS I was using for testing or something:
Coordinate Reference System:
  User input: unknown
  wkt:
PROJCRS["unknown",
    BASEGEOGCRS["GCS_unknown",
        DATUM["D_Unknown_based_on_Australian_Natl_S_Amer_1969_ellipsoid",
            ELLIPSOID["Australian_Natl_S_Amer_1969",6378160,298.25,
                LENGTHUNIT["metre",1,

and you can add SQL queries to `st_read` if that;s part of your motivation
for using a spatial file database:
pampa_srtm>300")
Reading query `select * from pts where pampa_srtm>300' from data source
`/tmp/olinda1_no.sqlite' using driver `SQLite'
Simple feature collection with 189 features and 1 field

Barry

On Thu, Mar 3, 2022 at 4:48 PM DUTRIEUX Loic <Loic.DUTRIEUX at ec.europa.eu>
wrote:

  
  


  


      

    

    

      
      

        


  

        

      Edzer Pebesma
    

    

      
      #
    

  


  

      
Loic,

I agree with Barry that there is a (logic) bug there, leading to a 
confusing error message. The problem seems the DBI driver failing to 
interpret a binary column ("blob") as wkb, which is needed when taking 
this path:


 > tb = dbReadTable(con, "olinda1")
 > head(tb)
   ogc_fid    id      cd_geocodi   tipo   cd_geocodb    nm_bair v014 
GEOMETRY
1       1 28801 260960005000001 URBANO 260960005020 Ouro Preto 1119 
blob[484 B]
2       2 28802 260960005000002 URBANO 260960005020 Ouro Preto 1267 
blob[532 B]
3       3 28803 260960005000003 URBANO 260960005020 Ouro Preto  557 
blob[276 B]
4       4 28804 260960005000004 URBANO 260960005020 Ouro Preto  709 
blob[468 B]
5       5 28805 260960005000005 URBANO 260960005020 Ouro Preto 1045 
blob[452 B]
6       6 28806 260960005000006 URBANO 260960005020 Ouro Preto  727 
blob[388 B]
 > st_as_sfc(tb$GEOMETRY)
wkbType: 30078980
Error in CPL_read_wkb(x, EWKB, spatialite) :
   unsupported wkbType dim in switch

so there seems to be something special with spatialite's WKB, which is 
not recognized by sf's WKB reader. Reading ?st_as_sfc:

 > st_as_sfc(tb$GEOMETRY, spatialite = TRUE)
Geometry set for 470 features
Geometry type: POLYGON
Dimension:     XY
Bounding box:  xmin: -34.91692 ymin: -8.044467 xmax: -34.82779 ymax: 
-7.954672
CRS:           NA
First 5 geometries:
POLYGON ((-34.86406 -7.992488, -34.86397 -7.992...
POLYGON ((-34.86129 -7.989947, -34.86087 -7.989...
POLYGON ((-34.85856 -7.992407, -34.85899 -7.992...
POLYGON ((-34.85752 -7.994827, -34.85722 -7.994...
POLYGON ((-34.8582 -7.997543, -34.858 -7.996937...
Warning message:
In CPL_crs_from_input(x) :
   GDAL Error 1: PROJ: proj_create_from_database: crs not found

This can no doubt be made more automatic (e.g. by adding a spatialite 
argument to st_read.DBIObject), but the the question remains why, when

x = st_read('/tmp/olinda1.sqlite')

works OK, that couldn't be used instead?

Many regards,

        
On 03/03/2022 19:19, Barry Rowlingson wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      DUTRIEUX Loic
    

    

      
      #
    

  


  

      
Thanks Edzer and Barry, this helps a lot.
For some reason I wrongly assumed that dsn= had to be a database connection for the query= argument to be valid.

The route st_read('/tmp/olinda1.sqlite', query=...) works perfectly fine, also for more complex query involving joins and multiple tables, therefore I will use that.
Happy to help if you consider that these features (reading spatialite via database connection, recognizing/parsing spatialite binary format, ...) deserves some attention.

Kind regards,
Lo?c

        

      
      
    

  
  


  


      

    

    
1 day later

    

      
      

        


  

        

      Micha Silver
    

    

      
      #
    

  


  

      
On 04/03/2022 13:36, Edzer Pebesma wrote:

            

      
      
      
    

        
I'm certainly no expert in WKB, but I did notice on the spatialite group 
list the following thread:

https://groups.google.com/g/spatialite-users/c/fJSlg6qGUfc

where Sandro explains that the spatialite "geom" column is NOT in 
standard WKB...

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Edzer Pebesma
    

    

      
      #
    

  


  

      
On 05/03/2022 16:29, Micha Silver wrote:

            

      
      
      
    

        
thanks - that's what sf takes care of when setting spatialite = TRUE in 
st_as_sfc():

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Micha Silver
    

    

      
      #
    

  


  

      
On 05/03/2022 17:52, Edzer Pebesma wrote:

            

      
      
      
    

        
Ah, got it. Thanks for clarifying