[gdal-dev] Strange errors with large vrt

[Even Rouault]

Le mardi 09 octobre 2012 19:08:33, yvecai a écrit :
> I'm trying to make a world-wide hillshade dataset filling SRTM void from
> Aster, and using Aster data only between 60°N and 72°N.
> I computed all 1x1° hillshade tiles, but now I'm struggling to work with
> a vrt.
> If I make a .vrt of all ~29'000 single 1x1° tiles, I don't have any data
> above 60°N, neither with Mapnik gdal datasource, nor gdal_translate, nor
> gdal2tiles.

Did you check that there was no error or warning reported when building the 
VRT ?

What do you mean by "I don't have any data" ? Do you mean that if you do an 
extraction from that VRT, you get no significant data ? What does gdalinfo 
report on the generated VRT ? Does it reported extent goes above 60° ? 

> But if I make a small 38-files .vrt around 60°N, gdal_translate with a
> -projwin works well.
> 
> Maybe someone can see a difference between hillshade2_N58E164.tif and
> hillshade2_N63E164.tif ? See gdalinfo below, I don't.

I don't see any difference... But all the files you provided in the archive are 
definitely odd : their first band is set to 0. Only the 2nd band (alpha band= 
has non uniform data. So that might explain what you saw with the huge .vrt

> Also, vrt files and some tiles can be found at
> http://www.pistes-nordiques.org/download/extract.zip (16MB).

No VRT files in that zip.