PG/OGR and OGDI issues - recent patch

[Markus Neteler]

Frank,

> >Briefly: The OGDI driver reports the VMAP0 'nam' field with
> >length of 64 chars, but delivers some strings in EURASIA
> >with > 64 chars. Subsequently the PG driver fails.

This was "fixed" by a patch to the PG driver, trimming the
values to the column length.

On Fri, Jul 09, 2004 at 12:41:18PM -0400, Frank Warmerdam wrote:
> I'm a little nervous about this fix.  First, I am not positive that the
> width is always set in a meaningful way for the PG driver.  Second, I am
> nervous about truncating data without at least warning the user.  Third,
> I feel the real bug is in the OGDI driver and should be fixed there.
> I sure wish the damn bugzilla was working so I could file this bug
> there and return to it.  Given that I can't, I think I will just apply
> your patch despite my misgivings.

Unfortunatelty this patch introduces a new problem:
The new cast fills the strings with trailing blanks so that e.g. 
MapServer now print all labels shifted (as it prints the white space
as well).

Example:
http://grass.itc.it/cgi-bin/mapserv?mode=browse&imgxy=250+250&imgext=10.132772+44.916805+12.500524+47.286189&imgbox=-1+-1+-1+-1&map=%2Fvar%2Fwww%2Fmap-script%2Fmapserverjs_vmap0_postgis.map&rawminx=10.132772&rawminy=44.916805&rawmaxx=12.500524&rawmaxy=47.286189&layer=coastl&layer=roadl&layer=railrdl&layer=watrcrsl&layer=inwatera&layer=builtupa&layer=oceansea&layer=elevation&mode=browse&zoom=1&mapsize=600+600&scale=&varscale=2071666.506602&mapxy=&map_resolution=100
(lake labels heavily shifted to west as white space appears right to the text)

Using a trim() maybe cures it. But it would be better not to allow
OGDI to send strings longer than the definition. Might this
be fixed in OGDI or the OGR/OGDI driver...

So you probably want to revert that change in ogrpgtablelayer.cpp
or add the SQL trim() function to get rid of the blanks introduced
by the typecast.

Example:
SELECT nam::varchar(6)||'end' from builtupa_eurnasia ;
 ?column?
-----------
 HAMMERend
 BATSFOend
 VADSO end
      ^-- undesired blank(s) inserted up to the length of the column

But:
SELECT trim(nam::varchar(6))||'end' from builtupa_eurnasia ;
 ?column?
-----------
 HAMMERend
 BATSFOend
 VADSOend
    ^-- looks much better

Best regards

 Markus Neteler

[warmerdam]

I have removed the previous change casting fields to the required length and
instead put in logic to trim overlong values and issue a debug message when
it occurs. 

I have also updated autotest/ogr_pg.py to have a test for this case to ensure
it works now and into the future. 

All seems well now.