[gdal-dev] Problem saving/reading inf GCPs values in VRT

[Calogero Mauceri]

Il 5/26/2015 11:28 AM, Even Rouault ha scritto:
> Le mardi 26 mai 2015 11:00:21, Calogero Mauceri a écrit :
>> Hi all,
>>
>> we are having problems saving/reading GCPs values in a VRT file when
>> those values are inf or nan. This problem is happening on Window
>> platform while it is working as expected in unix systems.
>>
>> We are setting the GCP values in a VRT dataset using the library
>> function GDALSetGCPs. The inf values get saved into the VRT XML as
>> "*1.#INF00000000E+000*" string
>>
>> <GCPList Projection="PROJCS["unnamed",
>>       GEOGCS["WGS 84",
>>           DATUM["WGS_1984",
>>               SPHEROID["WGS 84",6378137,298.257223563],
>>               TOWGS84[0,0,0,0,0,0,0]],
>>           PRIMEM["Greenwich",0],
>> UNIT["degree",0.0174532925199433]],
>>       PROJECTION["Orthographic"],
>> PARAMETER["latitude_of_origin",71.331088221],
>> PARAMETER["central_meridian",-156.63420306],
>>       PARAMETER["false_easting",0],
>>       PARAMETER["false_northing",0]]">
>>       <GCP Id="0" Pixel="0.0000" Line="0.0000" X="*1.#INF00000000E+000*"
>> Y="*1.#INF00000000E+000*" />
>>       <GCP Id="1" Pixel="363.0000" Line="0.0000" X="-3.463981152302E+004"
>> Y="3.246591123662E+004" />
>>       <GCP Id="2" Pixel="725.0000" Line="0.0000" X="-1.070555064325E+004"
>> Y="1.097994383230E+004" />
>>       <GCP Id="3" Pixel="1088.0000" Line="0.0000"
>> X="-5.461709489401E+003" Y="6.277974169290E+003" />
>>       <GCP Id="4" Pixel="1450.0000" Line="0.0000"
>> X="-3.134437286904E+003" Y="4.191819043427E+003" />
>>     </GCPList>
>>
>> when those values are read through the GDALGetGCPs library function,
>> then the inf values are set to 1.0.
>> Is this a bug in GDAL or are we setting the inf GCP values in the wrong
>> way?
>>
>> Note that on unix everything works fine, the inf values are saved as
>> "INF" strings in the VRT XML and they are properly converted when they
>> are read back.
> Calogero,
>
> Dealing with Inf or NaN serialized number on Windows require special cases.
> But I fail to see why you need to set GCPs at infinity or nan, and I'd bet the
> behaviour of the TPS or polynomial transforms will be pretty much undefined
> with such as value.
>
> Even

I agree with you Even, Inf and NaN values should not be saved as GCPs, 
but the way how they are treated in GDAL is very misleading, as invalid 
values become valid after saving and then reading them.
If you think that should not be fixed it should be at least reported in 
the documentation.

Calogero

-- 
Calogero Mauceri
Software Engineer

Applied Coherent Technology Corporation (ACT)
www.actgate.com