Sean Gillies

Intriguing!

@JonKing93 thanks for the report! I understand how the change could impact you. This was a deliberate change and we would like to keep it because it allows faster computation...

@JonKing93 these have been floats ever since GDAL 3.0, when float offsets and lengths for I/O was added to GDAL. The values are generally used for GDAL I/O and not...

@floriandeboissieu thank you for persisting! You are correct. The default behavior was to return ints as returned by `math.floor()`. In replacing `math.floor()` with `numpy.floor()` we accidentally changed the default behavior....

@mwtoews I'm not sure about this... GDAL isn't based on decimal math. And I wonder why the proposed change didn't break any tests? If the existing expectations are wrong, and...

@mwtoews this is too late for 1.4.0, but I don't want to rule it out for a future release. We've got workarounds for numerical imprecision everywhere, I'm unsure if this...

@mwtoews GDAL doesn't use decimal arithmetic, so any little changes we make in rasterio are lost when we use any GDAL methods. It's a little like making a promise that...

@groutr first of all: what impact would this have for rasters? Do we want two windows that are exactly adjacent, like tile windows, to intersect? Second: I think we may...

Thanks for the report @ZaraGi ! Yes, we need to define what the expected behavior is in this case.

Thanks for the report @roelofvandijkO ! I'll look into it. `num_threads` is passed directly to GDAL's warper and that's where the threading is managed. The PROJ error is interesting... maybe...