[gdal-dev] Motion: adopt RFC 87: Signed int8 data type for raster

[Even Rouault]

Hermann,

> I dare to say that, if we are following this path, maybe GDAL should 
> start marketing itself as a set of command line tools and not as a 
> library, given that breaking compatibility just because we can is not 
> something a library developer should do, IMHO.

I believe your words aren't reflecting the reality. We don't break 
things just to annoy people or because we can. We do it when we estimate 
it is in the best interest of the GDAL code base and its users, in the 
long term. There are adaptation costs for each side that can certainly 
felt as painful in the short/near term. That's life. There are past 
choices that turn out to be not ideal 15 years later down the road.

Regarding silent breakage, just have a look at 
https://gist.github.com/rouault/9d58bee0ceea111bd9913a03c9f13331 which 
shows an interesting change of behavior between C++03 and C++11. Or 
Python 2 vs Python 3 ( 2 / 3 = 0 vs 2 / 3 = 0.666667 ), or ... put it 
very long list here of software with long existence and that try to 
adapt themselves to remain relevant for their users and maintainable for 
their contributors ...

That said, https://github.com/OSGeo/gdal/pull/6722 should help 
addressing your concerns about silent breakage.

Even


-- 
http://www.spatialys.com
My software is free, but my time generally not.