How are you using the is_tiled property or tiled profile member? What if it was deprecated?

[sgillies]

I'm reconsidering the usefulness of a dataset's is_tiled property. When creating a new GeoTIFF, we must use the TILED=TRUE driver creation option because stripped TIFFs are the default. But GDAL, on which Rasterio is built, doesn't have a universal concept of whether a dataset is tiled or not. GDAL dataset or band objects don't have an "is tiled" attribute. Running gdalinfo on a GeoTIFF prints something like Band 1 Block=791x3 Type=Byte, ColorInterp=Red (from Rasterio's RGB.byte.tif) but does not say whether the dataset is tiled or stripped. GDAL knows about "block size" and that's it. Internally, GDAL rarely tests whether datasets are tiled or not, and when it does it is not much more than a comparison of block width to dataset width.

Whether a dataset is tiled or not, it's possible to iterate over chunks of it that are described by the block_shapes property. Does anyone have a use case where it matters if these chunks are tiles or strips?

[vincentsarago]

we do use is_tiled in our cog_validate script https://github.com/cogeotiff/rio-cogeo/blob/ff60e822a341f463d2be46a3e0444dd5c13ec705/rio_cogeo/cogeo.py#L467-L472

in GDAL, the cog validation script use a slightly different approach https://github.com/OSGeo/gdal/blob/master/swig/python/gdal-utils/osgeo_utils/samples/validate_cloud_optimized_geotiff.py#L197C13-L199 using the band.GetBlockSize which is what you mentioned in

it is not much more than a comparison of block width to dataset width.

I'm not really answering your question, but as a user of is_tiled I think it's a nice option to have 😄

[sgillies]

A related PR #3015 would change is_tiled to only check if dataset.block_shapes[0][1] == dataset.width. Which is what the GDAL COG tester does.