Is your feature request related to a problem? Please describe.kart import will convert tiles to cloud-optimised format during the import. Because the output is nondeterministic it makes an attempt to de-dupe these files based on the sourceOid from pointer files found at HEAD. The effect is that you can import the same dataset multiple times, and only changed files will actually be converted to cloud-optimised (identical tiles from the previous commit will be reused)
However, if the pointer files for existing tiles live on another ref rather than HEAD, it would be ideal if we could re-use those too.
Describe the solution you'd like
Add a kart import --find-tiles-at-all-refs flag to de-dupe tiles based on all refs.
Perhaps (probably based on whether it's very slow or not?) this could be the default behaviour. In case there are already relevant tiles to de-dupe this will be a considerable speedup, so it seems reasonable to accept at least a small performance loss in the general case (?)
Describe alternatives you've considered
The flag could take a set of refs patterns, but this might be overkill, e.g.
Is your feature request related to a problem? Please describe.
kart importwill convert tiles to cloud-optimised format during the import. Because the output is nondeterministic it makes an attempt to de-dupe these files based on thesourceOidfrom pointer files found at HEAD. The effect is that you can import the same dataset multiple times, and only changed files will actually be converted to cloud-optimised (identical tiles from the previous commit will be reused)However, if the pointer files for existing tiles live on another ref rather than HEAD, it would be ideal if we could re-use those too.
Describe the solution you'd like
Add a
kart import --find-tiles-at-all-refsflag to de-dupe tiles based on all refs.Perhaps (probably based on whether it's very slow or not?) this could be the default behaviour. In case there are already relevant tiles to de-dupe this will be a considerable speedup, so it seems reasonable to accept at least a small performance loss in the general case (?)
Describe alternatives you've considered The flag could take a set of refs patterns, but this might be overkill, e.g.