refactor(shared): move out Range class, memoize, code shuffle and type fixes

[clason]

• refactor: move main module to init.lua
• refactor(shared): move out Range class
• refactor(shared): inline and memoize match preparation
• refactor(shared): more memoize, type fixes
• refactor: consistent local api
• refactor(shared)!: drop support for specifying textobjects via regex
• refactor(shared): move {next,previous}_textobject to swap

Includes smaller code reshuffles and todos.

[clason]

@ribru17 sorry, this again conflicts (cosmetically) with your PR. Mostly it's just shuffling things around, though, so your logic changes should still apply pretty cleanly.

[clason]

Cursory tests seem to work fine, but I'd appreciate someone who actually uses this plugin to check for regressions.

[clason]

(The idea behind the refactors here and mentioned in the todos is to structure the modules so people will only have to pay for what they use, when they use it. It will also make it easier to work on each module without worrying about breaking other modules.)

[clason]

Scope textobject already fails on main; looks like the range class (or its call sites) needs a lot more nil guards.

[kiyoon]

@clason maybe the "lua match" matching of query names for the move feature should be removed as well. Maybe not many people use it and it just confuse them? Or make it as an option so that it is more clear that we opt-in.

But I would like to keep the "gradual movement" feature where you can set multiple query names. It is sometimes useful.

[clason]

Yes, I agree, globs are overkill (but lists are useful). Not sure I want to remove them in this PR, though. If we do, I would like to standardize on the signature, though (always take tables of strings) -- polyvalence is just painful.

[clason]

And thoughts on further splitting up shared?

• [x] shared.get_adjacent (very thin wrapper, can be inlined) and {next,previous}_textobject are only used in swap;
• [ ] shared.check_support and shared.available_textobjects can be moved to config to more cheaply check when setting up keymaps (this would require the mapped operations to silently fail on unsupported textobjects, which I think is the better UX anyway and is how every other textobject behaves).

(Alternative: the motion mappings themselves check and return false if not supported; the wrapper can then check the return value and call the original mapping instead: move(...) or ]m.)

[clason]

OK, I'm done with my murder refactoring spree for now. The next step is to rework the config (less metatable magic, more explicit) and the helpers for (buffer local) mapping setup, but that requires thought.

But before that, we should fix the many, many, regressions from master... (including getting the tests to pass again).

Would appreciate a thorough look, then I'll merge this and the CI PR as it's a strict improvement on current main.

[clason]

Tests pass now; will merge if nobody complains @kiyoon @ribru17

[ribru17]

Thanks a lot for this, can review now

[ribru17]

• shared.check_support and shared.available_textobjects can be moved to config to more cheaply check when setting up keymaps (this would require the mapped operations to silently fail on unsupported textobjects, which I think is the better UX anyway and is how every other textobject behaves).

(Alternative: the motion mappings themselves check and return false if not supported; the wrapper can then check the return value and call the original mapping instead: move(...) or ]m.)

Agree with all this

@ribru17 sorry, this again conflicts (cosmetically) with your PR. Mostly it's just shuffling things around, though, so your logic changes should still apply pretty cleanly.

No worries, I have accepted that that PR will come with lots of rebasing haha

[clason]

all yours, @ribru17 !

[ribru17]

lord have mercy on my soul...