[gdal-dev] gdal 1.8 tiff

[Even Rouault]

Le mercredi 11 mai 2011 18:51:07, Julien Malik a écrit :
> Hello list,
> 
> I had a bad surprise today using the new gdal 1.8 package available in
> ubuntugis-unstable :
> 
> When linked both to tiff v3 and gdal 1.8, the GTIFF driver is disabled
> with a warning appearing like :
> 
>     WARNING ! libtiff version mismatch : You're linking against libtiff
> 3.X but GDAL has been compiled against libtiff >= 4.0.0
> 
> I understand from frmts/gtiff/geotiff.cpp that this is voluntary (though
> the warning does not say it actually disables the GTiff driver).

I'm a bit frustrated... I put this check to avoid bug reports actually ;-) Too 
many people get hurt because of a version mismatch and have a hard tried 
figuring what goes wrong.

> 
> In my case (building the Orfeo Toolbox library, which has a dependency
> needing TIFF and GeoTiff, but not gdal), it has the undesirable
> consequence of breaking TIFF support for our ubuntu packages based on
> the ubuntugis-unstable ppa
> (https://launchpad.net/~otb/+archive/orfeotoolbox-nightly and
> https://launchpad.net/~otb/+archive/orfeotoolbox-stable-ubuntugis).
> 
> The issue has already been discussed on this list, but just to recall :
> we had a solution providing usable executables with TIFF support, even
> though they were linked to the system libtiff and libgeotiff (it
> involved ensuring the proper link order between '-lgdal' and '-lgeotiff').
> 
> That's why I think disabling the driver completely is a little hard.
> In my case it represents a major regression, and I don't have a lot of
> ideas to work around it...
> 
> 
> Is it possible to *not* disable the driver, and keep the warning ?

If your question is "without changing the code ?", the answer is no.

If Debian/Ubuntu goes into the way of using versionned symbols, they can just 
apply a local trivial patch to disable this check. My rationale is that people 
that want to mix both libtiff 3 and libtiff 4 are considered to be sufficient 
advanced users to make patches to undo the sanity checks if they really know 
what they are doing. IMHO, the default should be to stick to safe settings.

But perhaps I did go too far and I'm open to suggestions. Changing the 
behaviour of this check is trivial. The difficult part is to figure out the 
expected behaviour and I just don't know what would be the option that would 
satisfy most/all people. One possibility would be to disable the driver unless 
an environment variable GDAL_ACCEPT_MIXED_LIBTIFF_VERSIONS is defined  ?? But 
that seems to a bit too convoluted. Or just emit the warning and still enables 
the driver, considering that the warning is enough and that if people get 
crashes, they would know why. I'm interested by opinion from other people.

Best regards,

Even
> 
> 
> Regards,
> Julien
> 
> NB: I filed a ticket in our bug tracker
> (http://bugs.orfeo-toolbox.org/view.php?id=296), feel free to leave
> comments there.
> 
> 
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/gdal-dev