Panu Lahtinen

> Oh, I didn't know that different area possibility! Actually I "knew" that existed because I had tested the temporal RGB stuff also with polar orbiter data, just didn't realise...

Yes, that's what I was thinking.

> The downside of the second approach is that it makes it harder to experiment with different ways of combining time-steps in different RGBs or other composites, as this would...

I copied the two lines to `noaa21.tle` file, pointed `PPP_CONFIG_DIR` to directory of my `platforms.txt` where I added `NOAA-21 54234`. After this ```python from pyorbital import tlefile tle1 = tlefile.read('noaa-21',...

This is all the properties that can be cached without breaking the existing tests, and also giving me so far the best performance. Some of the remaining properties cause failure...

> `read_module` is actually no real fixture, it's _defined_ in this helper function, which will be called with the real test-functions: > > https://github.com/h5netcdf/h5netcdf/blob/2771793b27588026bf11b81b9cd77aabec59e187/h5netcdf/tests/test_h5netcdf.py#L446-L449 Oooh, indeed, thanks!

Thank you @kmuehlbauer ! I can get the legacy API tests to pass while caching `Dimension._hd5s` and `Dimension._Dimension._group` for `mode="r"` (and again gain few seconds in processing time, now down...

I might be misunderstanding a lot, but I'm not sure the `test_no_circular_references()` tests for _circular_ references, just that there should only be one reference to `ds` while the file is...

> i think this is also the source of test failures for xarray. > > We need to make it a weakref somehow. What are you referring to in particular...

I think I'm ready with this stage of optimization. I've adapted the test script from https://github.com/h5netcdf/h5netcdf/issues/195#issuecomment-1336182157 to use caching benefits of `mode="r"`. The file is also closed between the tests...