[Qgis-developer] Re: Raster providers

[Radim Blazek]

On Tue, Jan 18, 2011 at 9:19 AM, Marco Hugentobler
<marco.hugentobler at sourcepole.ch> wrote:

> oops, that should be 'Agreed, int to double conversion could be problematic'

Why do you consider it problematic, because it occupies too much space?
Regarding precision, int to double is OK, I believe, but long to
double is not because double has only 52bits to store mantissa. GDAL
however does not define long,
the longest integer is GDT_Int32

> What about an approach where the out-of-bound values are assigned the last
> value only if reprojection is used (and if the layer does not have a null
> value)?

Still the problem remains for projected data.

I think that usually with common data and wider data type like int,
float and double the max value is never used except for some special
values like null. The problem occurs
with shorter types like byte and short. So I suggest:

source_type  -> qgis_type
byte -> byte (if null value is defined) / short (if null value is not defined)
short -> short  (if null value is defined) / int (if null value is not defined)
int -> int, if null value is not defined use max value for null
float -> float, use NaN for null values
double -> double, use NaN for null values

This is still not 100% solution, but it should cover most common cases.
The only code complication is, that we need a new srcDataType()
provider method,
to return the true data source type.

Radim

> This would be quite similar to your current approach, except that people will
> have the appropriate raster appearance if they don't use reproj. And if they
> do, they have an undesired border, but still the right colors in the raster
> area.
>
> Regards,
> Marco
>
> Am Montag, 17. Januar 2011, um 17.34:22 schrieb Radim Blazek:
>> On Mon, Jan 17, 2011 at 10:06 AM, Marco Hugentobler
>> <marco.hugentobler at sourcepole.ch> > Using NaN sounds like a good idea
>> and Qt has platform independent support for
>>
>> > it (qIsNan & co.).
>> > All other solutions I can think of seem to be more complicated (e.g.
>> > force a transparency value only if raster is reprojected).
>> >
>> >>float + NaN for byte/int?
>> >>
>> > This is not clear to me. Could you explain your approach for byte/int?
>>
>> I thought that if data type of the source is integer the provider
>> could represent them as floating point. Byte can be represented as
>> integer. Bad solution however.
>>
>> Radim
>
>
> --
> Dr. Marco Hugentobler
> Sourcepole -  Linux & Open Source Solutions
> Churerstrasse 22, CH-8808 Pfäffikon SZ, Switzerland
> marco.hugentobler at sourcepole.ch http://www.sourcepole.ch
> Technical Advisor QGIS Project Steering Committee
>