[gdal-dev] VRTComplexSource with a LUT, proposal

[Even Rouault]

Le dimanche 20 mai 2012 16:07:07, Saâd HESSANE a écrit :
> Hi Even,
> 
> The main goal of this proposal is to built using "gdalbuildvrt" VRT files
> containing LUT or filters.
> 
> The patch that I send allow build VRT with LUT by specifying the -lutin and
> -lutout parameters.
> Ex :
>   gdalbuildvrt -lutin "0 1 2" -lutout "4 5 6" test.vrt test.tif
> 
> We can discuss the fact that define two parameters (-lutin and -lutout) or
> just one parameter (-lut "0,4 1,5 2,6"). This is not a problem.

Ah, your patch is more extensive than what I thought (I imagined only the 
SetLUT() function to be in it).

I see that your patch changes the definition of a function in the C API 
(VRTAddComplexSource). Up to now, the GDAL C API has been set in stone. This 
might change in GDAL 2.0 though, but I'm going to start a specific thread on 
this to have global feedback from the GDAL community.

But even if we are OK to change the C API, I begin to think that this 
VRTAddComplexSource() function becomes ugly with so many parameters. Perhaps 
we should add a char** papszOptions to provide additional options without 
breaking it each time. Or I'm wondering if the C API is really the best option 
for doing such complex things, and using the C++ API wouldn't be just better.

> 
> 
> 2012/5/19 Even Rouault <even.rouault at mines-paris.org>
> 
> > Le samedi 19 mai 2012 18:58:05, Saâd HESSANE a écrit :
> > > Hi Even and thank you for the quick response.
> > > 
> > > You should even not consider the VRTComplexSource class to be in the
> > 
> > public
> > 
> > > > API, so the visibility of its members is not significant. And playing
> > > > with them
> > > > from the outside isn't recommanded at all.
> > > 
> > > The fact that you don't consider the class in the public API does not
> > > excuse the fact that this is a mistake encapsulation :)
> > 
> > I agree the encapsulation isn't ideal, but unless I'm wrong, this class
> > is *not* in the public API. It is not marked as CPL_DLL exported, so on
> > Windows,
> > you should not be able to access it from the outside of the GDAL lib. On
> > Unix/Linux, as, by default (unless GDAL is configured with
> > --hide-internal- symbols), all symbols are exported, you can technically
> > use it however.
> > 
> > > And apart from that, nothing prevents me to use a special driver to do
> > > specific things that are not directly provided by the public API.
> > > Currently I need to build a VRT, and the VRT driver are fine for that.
> > 
> > It's
> > 
> > > dirrectly use in the gdalbuildvrt utility for example. If the VRT
> > > plugin
> > 
> > is
> > 
> > > not safe to use, the solution is to correct it.
> > 
> > There's always a trade off between exposing API and increasing
> > maintenance burden. The more API you expose, the more difficult it is to
> > make changes afterwards.
> > 
> > > To read a VRT file outside from gdal I need to parse an XML file, so I
> > 
> > have
> > 
> > > to use another dependency (like xerces) to do just a small think.
> > 
> > You don't need another dependency. You can use the CPL minixml API (see
> > cpl_minixml.h) that is used by the VRT driver itself for example.
> > 
> > > Another argument is the VRTKernelFiltredSource. To set a kernel filtre
> > > we don't have to set the attributes nKernelSize, padfKernelCoefs and
> > > bNormalized, because we can't (the attributes are protected). But the
> > > API offers a setKernel that do exactly the same think that I hope the
> > > setLUT method do in the VRTComplexSource.
> > 
> > If you can come with a patch, I'll consider including it, but I still
> > believe
> > you should not rely on that.