Hey all, I work with Kurt Schwehr, and we've been debugging failures where an explicit mask isn't preserved through a CreateCopy operation to create a COG through the GTiff driver at HEAD.
This failure only started occurring after https://github.com/OSGeo/gdal/commit/3bdc81a205bad8342688873e4ed5716f1208e8f5 . After the CreateCopy, when we read back the mask, the mask is expanded from the 1-bit mask to the 8-bit mask inside of gtiffoddbitsband.h. However, the mask is not a 1-bit mask when I add logging here; instead, it's the original 8-bit mask (with 255s and 0s). So when it gets expanded to 8-bits, it's no longer the original mask: E.g., 255 0 255 255 becomes 255 255 ... 255 255 0 0 ... 0 0 255 255 ... So my guess is that somewhere along the line we either copy the mask without turning it back into a bitmask, or we lose metadata tracking that the mask is encoded a certain way. Does anything obvious jump out to folks here? The issue *does not* occur when either COPY_SRC_OVERVIEWS="YES" is removed or INTERLEAVE is set to PIXEL (it only fails on band interleaving). Here's a gist containing an example test that fails at head but previously succeeded: https://gist.github.com/orglofch/40cf4d9136978b0384141cc74d1039b6 . Thanks, Owen
_______________________________________________ gdal-dev mailing list gdal-dev@lists.osgeo.org https://lists.osgeo.org/mailman/listinfo/gdal-dev
