Add alias completion

[knezi]

Partially fixes #32.

I wasn't able to add tests. I feel like it could be added in tests/unit/completion/test_models.py just like bind command (test_bind_completion and others).

Or in test_completer.py by adding more parameters in parametrized test_update_completion.

Can you give me a hint?

[rcorre]

I think test_models would be the appropriate place to add tests.

[rcorre]

This works, though I wonder if it's overkill and we should just use config.cache['aliases'][before_cursor[0]]?

[toofar]

+1, both of looking stuff up in objects.commands and config.cache['aliases'] are an order of magnitude faster than parsing a command line. And that's just what the command parser does to resolve aliases.

@knezi you should be able to plop a config_stub.val.aliases = {"openalias": "open"}, or whatever, in test_update_completion and add some more parameterized test entries.

[knezi]

Thanks for the feedback, I've updated the commit accordingly. The failed tests doesn't seem to be caused by this commit.

[rcorre]

You should only need one dict lookup here, something like:

Suggested change

	cmdname = before_cursor[0]
	if before_cursor[0] in config.cache['aliases']:
	cmdname = config.cache['aliases'][before_cursor[0]].split(
	maxsplit=1)[0]
	cmdname = before_cursor[0]
	alias = config.cache['aliases'].get(cmdname)
	if alias:
	cmdname = alias.split(maxsplit=1)[0]

[knezi]

Actually, this doesn't work for obscure cases like alias ba for bind <ctrl-x>. If you run ba |, it won't complete commands as it expects key short cut first.

Other examples are set if you want to quickly set some particular settings. Or possibly
config-dict-add and hint, but it's not applicable for those as the completion is not done for these commands.

So I could replace before_cursor[0] with the entire alias split on spaces, or already pass expanded alias to _get_new_completion. What do you think?

[The-Compiler]

I vaguely remember cases like those are why I removed alias completion again (b703028, #358) after initially implementing it, some 9 years ago.

I don't remember the details, but this seemed to be surprisingly complex to get right. Maybe something changed about that with the refactorings done since then, but maybe not. There are some ideas floating around regarding passing the command-line around internally as some kind of token objects rather than a string, which I suppose might make it easier.