Dion Häfner

Let users supply key names before deployment?

2

Currently, key names are read from the database. Alternatively, we could require users to supply both a database *and* the associated keys. Pro: - API spec can include key names,...

Use rio-cogeo internally

7

Currently, we have our own functions for optimizing and validating COG, which are more or less copied from `rio-cogeo`. I did this because `rio-cogeo` did not feel quite mature enough...

Write a client library

3

Terracotta would only be a complete A-Z solution if we also provided a client library in Javascript to do some of the heavier lifting. I don't see this happen anytime...

We do have logging support, but use it pretty sparingly. Possible enhancements include: - Print SQL queries for SQL-based drivers - Forward some information from GDAL, like total number of...

This could imply storing hashes, file sizes, or dates.

When serving heterogeneous datasets (e.g. data with a `type` key), if could be very useful to be able to retrieve all unique key values. Examples: 1. Get all unique keys...

Support non-linear rescaling

1

Such as conversions between different color spaces

Internal state modified after catching exception in _create_child_variable

4

Example script: ```python import h5netcdf import numpy as np with h5netcdf.File("test.nc", "w") as f: try: f.create_variable("test", ("x", "y"), data=np.ones((10, 10), dtype="bool")) except: pass print(f.dimensions) ``` Output: ``` $ python bug.py...

I am the maintainer of [`pyhpc-benchmarks`](https://github.com/dionhaefner/pyhpc-benchmarks), a repository that compares the performance of various JIT compilers and ML frameworks on scientific computing workloads (mostly finite difference stencils from an ocean...