Depend on official homebrew formulas for dependencies

[sebroeder]

I worked on the "low-hanging fruit" of #115, replacing bundled dependencies with its Homebrew counterparts where official formulas are available.

Work remaining:

• Create a Homebrew formula for libvterm because there isn't one yet. It would be helpful if a release of libvterm could be tagged for this purpose.
• Migrate luajit and luarocks dependencies to formulas, too.
  luarocks is not available as a separate formula in Homebrew but the lua formula installs it by default. I could not get the bundled luarocks to build with the luajit from Homebrew. We could depend on the lua formula to get luarocks but that seems not very elegant. I would prefer if we could help out with Add support to luarocks to luajit, code adapted from lua formula Homebrew/legacy-homebrew#41298 to bundle luarocks with luajit the same way it is done for lua.
• Remove the remaining logic in the formula that deals with building the dependencies, once the two tasks above are done.

Some official formulas are a little behind the bundled versions and could be bumped.

[sebroeder]

Any feedback on this? It the general direction right? Should I keep working on this?

[jszakmeister]

Honestly, I think it depends on Neovim's situation. @tarruda Have the dependencies stabilized? To me, 0.1 doesn't signify that we're ready to do that just yet, but I'm really not sure we sit on the spectrum right now.

[sebroeder]

This is a reasonable point to consider, thanks @jszakmeister. Even if there are some dependencies that are still in flux I would propose to gradually move to external formula dependencies, to make the eventual final migration easier. This would also make it more transparent which dependencies are already considered stable and which ones are not.

[jszakmeister]

Even if there are some dependencies that are still in flux I would propose to gradually move to external formula dependencies, to make the eventual final migration easier.

Sure, I'm just not certain which ones those would be. :-) If I had to guess, I'd say it's likely that LuaJIT, jemalloc, libuv, and msgpack likely fit the bill.

LuaJIT is probably the more painful one to deal with, since it's more of a build dependency at the moment and we want our test infrastructure installed. We should probably look to have it use a separate rock tree to avoid polluting the user's rock tree. That would all have to happen in Neovim proper.

The others should be much more straight-forward, assuming they're stable.

This would also make it more transparent which dependencies are already considered stable and which ones are not.

I don't think the Homebrew formula is really the place to document this as much as the third-party build infrastructure in Neovim itself. I don't think anyone on the core team is going to be looking at the Homebrew formula to answer that question. ;-)

[tarruda]

I don't think the Homebrew formula is really the place to document this as much as the third-party build infrastructure in Neovim itself. I don't think anyone on the core team is going to be looking at the Homebrew formula to answer that question. ;-)

Agreed

As for the dependencies, I'm sure there will be changes before 1.0

[parkr]

Would love to see this! 👍

[dholm]

@sebroeder You can grab whatever changes you are missing from #129.

[galexite]

I've since submitted libvterm to the Homebrew core formulae – libvterm therefore can be required as a formulae dependency.

[equal-l2]

Any progress about luajit and luarocks?

[justinmk]

continued at #195

Thanks @sebroeder for getting this started, and @galexite for submitting libvterm.