Implement command-line tool interface

[simonw]

The first version needs to take one or more file names or URLs, then generate and deploy an app to Now. It will assume you already have the now command installed and configured.

[simonw]

Simplest implementation will be to create a temporary directory somewhere, copy in a Dockerfile and the databases and run “now” in it.

Ideally I can use symlinks rather than copying potentially large database files around.

[simonw]

I’m going to have a single command-line app that does everything.

Name to be decided - options include dataset, stateless, datasite (I quite like that - it reflects SQLite and the fact that you create a website)

[simonw]

datasite .

• starts web app in current directory, serving all DB files

datasite . -p 8001

• serves on custom port

datasite blah.db blah2.db

• serves specified files

You can’t specify more than one directory. You can specify as many files as you like. If you specify two files with different oaths but the same name then they must be accessed by hash.

datasite publish .

• publishes current directory to the internet! Uses now by default, if it detects it on your path. Other publishers will be eventually added as plugins.

datasite publish http://path-to-db.db

• publishes a DB available at a URL. Works by constructing the Dockerfile with wget calls in it.

datasite blah.db -m metadata.json

If you specify a directory it looks for metadata.json in that directory. Otherwise you can pass an explicit metadata file oath with -m or —metadata

[simonw]

Another potential name: datapi

[simonw]

I’m going to use Click for this

http://nvie.com/posts/writing-a-cli-in-python-in-under-60-seconds/
https://kushaldas.in/posts/building-command-line-tools-in-python-with-click.html

[simonw]

Here’s how to make the “serve” subcommand the default if it is called with no arguments:

@click.group(invoke_without_command=True)
def serve():
    # ...

[simonw]

Deploys to Now aren't working at the moment - they aren't showing the uploaded databases, because I've broken the path handling somehow. I need to do a bit more work here.

[simonw]

http://the-hitchhikers-guide-to-packaging.readthedocs.io/en/latest/quickstart.html describes how to package this for PyPI

[simonw]

To simplify things a bit, I'm going to require that every database is explicitly listed in the command line. I won't support "serve everything in this directory" for the moment.

[simonw]

... I tried that, I don't like it. I'm going to bring back "directory serving" by allowing you to pass a directory as an argument to datasite (including datasite .). I may even make . the default if you don't provide anything at all.

[simonw]

I'm happy with this now that I've implemented the publish command in #26