Should we speed-up xcdat import?

[pochedls]

It feels like xcdat is slow on import (i.e., import xcdat), though I recognize it only takes ~6s. This is much slower than other libraries (xarray, pandas, and scipy clock in under 1s). They are probably loading sub-modules or dependencies on demand. Some questions:

• Do others also feel xcdat is a little too sluggish on import?
• Maybe we should profile this to see what is causing the slowdown? Are we importing the lowest level of modules needed?
• Should we defer imports of dependencies until we need them? Or use lazy_loader?

Thoughts?

[tomvothecoder]

Do others also feel xcdat is a little too sluggish on import?

Thanks for opening this issue. I've noticed xCDAT loads slow too (~6 seems slow to me).

Maybe we should profile this to see what is causing the slowdown? Are we importing the lowest level of modules needed?

This is happening because running import xcdat imports everything in __init__.py (here).

As you suggested, we'll need to profile the imports to pinpoint the exact packages that are slow. In my quick-test I noticed import xesmf and import xgcm are both slow with the first call (subsequent calls are fast due to caching I think).

Should we defer imports of dependencies until we need them? Or use lazy_loader?

This would be great to improve the speed of import xcdat, if possible.

[tomvothecoder]

This is closed by #810