Implement XDG Base Directory specification

[JustAnotherArchivist]

ia will attempt the following paths in this order, using the first existing file:

• ${XDG_CONFIG_HOME}/internetarchive/ia.ini (with XDG_CONFIG_HOME defaulting to ${HOME}/.config if not set or invalid)
• ~/.config/ia.ini
• ~/.ia

When none of these paths exist on creating a config file (ia configure), the first one is used and directories are created as necessary.

Closes #400

[JustAnotherArchivist]

A note on os.environ['HOME'] vs os.expanduser('~'): the XDG specification requires using $HOME. os.expanduser falls back to using the password database or the root directory if $HOME doesn't exist and is therefore not fit for purpose. In practice, this shouldn't matter as POSIX requires the presence of that environment variable, but I feel like it was worth mentioning anyway.

[JustAnotherArchivist]

For some reason, $HOME apparently isn't set in Travis CI. Lovely...
Not sure what the reason for that is. Smells like a bug on Travis's side.

I've successfully tested this code locally with a POSIX-compliant environment in all scenarios I could think of.

[JustAnotherArchivist]

Rethinking this, I suppose ia should also work on Windows, where $HOME doesn't exist as far as I know. Perhaps using ~ is better after all; on any properly POSIX-compliant system, it will still behave according to the specification since expanduser uses $HOME if set. This would also work around the Travis CI bug. @jjjake, any thoughts on this?

[JustAnotherArchivist]

I made that change from $HOME to ~ as I can't really think of a good reason not to do that. The specification doesn't say anything about what should happen if $HOME isn't present, presumably because it assumes a POSIX-compliant system (without mentioning it). And on such systems, the new code will still behave according to the specification because Python's expansion uses $HOME if present on Unix and only falls back to the password database if not; on Windows, $HOME is useless anyway.

I would still like to figure out why the build failed though. I wrote to Travis CI, and they confirmed that $HOME works correctly in an independent test build. I'll probably create a separate PR to play with that.

Also, I still want to add some tests here to ensure that a #399 situation can't happen again. So this isn't ready for merging yet.

[JustAnotherArchivist]

Tests are succeeding except on Python 2.7 because tempfile.TemporaryDirectory doesn't exist there. I could try to reimplement that, but Py2 is EOL and ia is already broken on it (#385), so I'm not sure that's worth the effort. @jjjake, please advise.

[JustAnotherArchivist]

I took a quick look at the implementation of tempfile.TemporaryDirectory, and it's more complex than I thought. If dropping Py2 support is not yet an option, skipping these tests on Py2 would be another possibility.

[jjjake]

I understand PY2 is EOL, but it's still widely used in reality and something I would like to support as long as we're able to. Any changes made in support of XDG_CONFIG_HOME should not break ia configure, etc., in PY2.

[JustAnotherArchivist]

Alright, I understand. To be clear, it's only the tests that are held back by this; the actual code should not be Py3-specific in any way.

I'll try to write a minimal crutch for Py2 that emulates the key part of TemporaryDirectory. I have no desire to reimplement its complex error handling though – a lot of things in that area changed between Py2 and Py3 and I haven't written code for Py2 in years... But it should be good enough to get the test suite running on CI at least.

[JustAnotherArchivist]

I realised today that write_config_file was broken on Python 2 for the same reason as the tests were previously (lack of exist_ok kwarg), but because there were no tests for that method at all, it didn't trip anything. I refactored that code slightly to decouple the IA login from the config file writing and added tests for all of that as well.

This should now be ready for review and merging. :-)

[jjjake]

Thanks again, @JustAnotherArchivist!