Packer v2 Rewrite

[wbthomason]

This is a WIP PR to complete the first stage of this
roadmap,
at least through the load function.

Current progress:

• Add some initial TODOs
• Put configuration into its own module
• WIP main module rewrite
• Implement specification flattening
• Add ensure_table function
• WIP flattening iterator for plugin specs
• Minor code cleanup, preparing to restart work
• Simplify headless check
• Simplify ensure_table
• Use new Lua commands API for Packer commands
• Use new Lua autocommands API for PackerComplete
• WIP moving to runtime-handlers implementation
• Add path lib
• Make profile lib from compile functions
• Always ensure dependencies
• Use path lib in config
• Signal existence of profiling threshold config var
• Add hooks for modules that need to know about updated config values
• WIP everything-as-handlers-no-compile approach
• Belay changing load for now
• Add proper timed_load and rename old one to timed_packadd
• Cleanup log module use a bit
• WIP handler definitions
• Add note about feature request from lewis6991
• First draft of rewrite roadmap

The next steps are:

• Completing the keyword handler set
• Writing the load() function
• Presumably fixing some bugs
• Potentially some re-engineering to minimize the startup cost of now always needing to require the packer module and your plugin specs.

Contributions to this PR are welcome!

[wbthomason]

Tagging @akinsho, @EdenEast, anyone else interested in contributing to the rewrite.

[AckslD]

Generally this looks great!

Regarding redundancy, there's a lot of code-duplication in packer.update and packer.sync. Maybe there are reasons for this but if things are structured well, it seems like sync could just be something like:

packer.sync = function(..)
  packer.clean()
  packer.update(...)
  packer.compile()
end

[wbthomason]

That's the idea! I just haven't gotten to that part of the rewrite yet.

[akinsho]

@wbthomason thanks for opening it, looks good 👍🏿. I was thinking I'd raise some PRs against this to help move it on, but will probably start with smaller things like adding Emmy Lua annotations to as many places as possible. I personally heavily rely on those when writing Lua, so would be nice if the types in the code base were clearer.

I've also been using the lazy loading pattern popularized by TJ in https://github.com/tjdevries/lazy.nvim and thought I might add it here? It defers the requiring of modules till they are actually indexed, so you can require all modules at the top of a file, but they won't actually be required till they are used. You can then combine this with sumneko's @module syntax to keep the type information, e.g.

local lazy = require('packer.lazy')
--- @module 'packer.config'
local config = lazy.require('packer.config')

Primarily because you've talked about reducing the cost of requiring packer, so this should help.

[wbthomason]

Oh, brilliant! That looks quite useful. Separately, more annotations is also a great way to start - that's a stated goal of the rewrite and will make working with the codebase substantially nicer. Thanks!