Sean Gillies

The history of os.register_at_fork is an interesting read: https://bugs.python.org/issue16500.

@skogler I have confirmed that the test fails without the patches. Thank you for putting this together. @rouault can I ask you to consult on this? Is this the sensible...

@rouault thanks! @skogler I modified your test to call `multiprocessing.set_start_method("spawn")` before the new Process is created (see https://docs.python.org/3/library/multiprocessing.html#multiprocessing.set_start_method) and no longer see a deadlock. Since Python 3.7 we can pass...

If we fork before rasterio.env.local._env is initialized, I don't see a problem. `MemoryFile.open` has a decorator that initializes the GDAL environment (including driver registration): https://github.com/rasterio/rasterio/blob/master/rasterio/env.py#L385. I think we could raise...

@karoleks4 rasterio 1.2.6 and GDAL 3.2.2 are a bit old. Do you see the same thing with rasterio 1.2.10 (or 1.3b1) and GDAL 3.4 or 3.5? GDAL has a number...

@karoleks4 @DmitryAVasin thank you for bringing this up. I'm working on reproducing this and figuring out what's up. I observe that memory usage increases less when I decimate by a...

@daniel-j-h these options were implemented naively and it will be a fair amount of work to change this. For what it's worth, the click project doesn't see this as a...

@groutr I prefer the second approach. A simple sorting should be fine. Instead of sorting inside sample_gen, what would you think about making a separate sorter? Then the function remains...

This issue will be resolved along with #1990.

@recursix can you clarify: do any of the three things you enumerated solve the problem by itself? Or are all 3 required? This is almost certainly a GDAL bug because...