[Skyfield](https://rhodesmill.org/skyfield/) is a Python library for astronomical computations. It depends on various data files to accurately compute moon phases, planet positions, etc.
Several issues are raised by these data files:
* If they're not found in the path of the ``Loader``, they're **downloaded at runtime**. Depending on the archive you're requesting, some files might be very large, causing a long delay (directly related to your network bandwidth). In the case of a web server app, you'd cause a timeout on client's end.
* They come mainly from 3 sources: the USNO (US Navy), Paris (Meudon) Observatory, and NASA JPL. **If one of them is temporarily unavailable**, you couldn't perform any computation.
* In some countries, or behind some filtering proxies, the USNO is considered as a military website, and thus is **blocked**.
* These files have **an expiration date** (in a more or less distant future). As a consequence, even if the files are already downloaded in the right path, at each runtime you could possibly have to download one or more files before making any computation using them.
* Providing at least the most common of these assets in Python Package.
* Make regular releases to refresh the files before they expire.
* Provide a warning / logging mechanism when the files are about to expire (or when they are outdated) to still allow you to compute things with the loaded assets, but being informed you need to upgrade.
This way, you could **install or upgrade** this data package via ``pip``.
Once all the files are on your disk space, you can instantiate your ``skyfield`` loader pointing at their path, without having to worry about anything.
Whenever a file contained in the catalog has expired, you're going to receive a warning when loading the `skyfield-data` path:
```python
>>> from skyfield_data import get_skyfield_data_path
>>> from skyfield.api import Loader
>>> load = Loader(get_skyfield_data_path())
/home/[redacted]/skyfield_data/expirations.py:25: RuntimeWarning: The file de421.bsp has expired. Please upgrade your version of `skyfield-data` or expect computation errors
RuntimeWarning
```
By default, the loading isn't blocked, but it's strongly recommended to upgrade to a more recent version, to make sure you're not going to make wrong astronomical computations.
By default, the ``RuntimeWarning`` is raised when the file **has** expired. You may want to be aware of this warning **in advance**, that is to say a few days or weeks before, in order to eventually upgrade your version of ``skyfield-data``.
In order to trigger this warning, you can use the ``expiration_limit`` argument, like this:
```python
>>> from skyfield_data import get_skyfield_data_path
/home/[redacted]/skyfield_data/expirations.py:25: RuntimeWarning: The file de421.bsp would expire in less than 30 days. Please upgrade your version of `skyfield-data` or expect computation errors
RuntimeWarning
```
**Note:** The ``expiration_limit`` argument should be a positive integer (or zero).
In order to be able to run the `download.py` script, we recommend to run it **from a virtualenv** where you'd have installed the "dev" dependencies, using:
*Important:* This project is, and should stay compatible with Python 2.6, 2.7 and Python 3.5+ up to 3.8, to keep the same Python compatibility that `skyfield` has.
### Hacking
Improving or fixing `skyfield-data` will require you to have at least a virtualenv with `tox` installed on it.
We'll ask you to add tests along your patch, to make sure that no regression or bug would be introduced by your patch or further ones.
To make a quick'n'dirty test, inside your "tox-ready" virtualenv, run:
```sh
make test
```
to launch the Python 2.7 and Python 3.5+ test jobs.
If you want to test your branch against Python 2.6, you'll have to setup a Python 2.6-ready tox environment, by doing something similar to:
```sh
sudo apt install python2.6 python2.6-dev # dev headers to compile numpy
mkvirtualenv TOX26 --python=`which python2.6` # You will activate this venv with `workon TOX26`
pip install tox
tox -c tox-py26.ini
```
**Known issues**: on Ubuntu, you may be unable to build numpy at this point, due to misplaced C header files in your system. I've had hard times on Ubuntu, but your mileage may vary.
If you don't want to or can't install Python 2.6 requirements, you'll have to rely on the Online CI jobs by Circle-CI.
