[feature] plugins

Was thinking of some stuff I’d like bankai to do, and then realized it might get too bloaty. Instead having a plugin system might perhaps help. Is it great? Nah. But it’s probably the lesser of a bunch of evils.

I think we could probably learn from browserify’s plugin architecture because it’s super solid. I don’t think transforms make a lot of sense for us here because it could refer to either JS, CSS or HTML which already have specific transforms available (html not so much, but yeah).

I was thinking in package.json people could do something like:

{
  "bankai": {
    "plugin": [
      "server-render/plugin"
    ]
  }
}

Or from the CLI:

$ bankai -p server-render/plugin

plugin examples

• compile all routes in the choo router to static html
• module split on a per-route basis
• inline all JS and CSS into the HTML (goes nicely with the two points above)
• perhaps doing performance budgets is like cool (e.g. https://github.com/yoshuawuyts/bankai/issues/126)
• optimize and or compile particular assets (e.g. https://github.com/yoshuawuyts/bankai/issues/104)
• use packages like purifycss which require knowledge of multiple resource types (e.g. JS and CSS in this case)

todo

• figure out if this is actually the best way to approach the problem
• come up with a JS API for the plugins
• create example implementation for some existing things

cc/ @toddself @kristoferjoseph think this might be of interest for y’all; input would be v welcome!

[toddself]

to summarize from irc: i’m not sure bankai needs a plugin system. Sometimes bloat on a developer tool is OK since it takes away the paralysis of choice (eg: standard, tap). Install size is partially irrelevant (or you wouldn’t use it, or you can install it globally), and if you aren’t using something that comes with it, you’re not increasing my cognitive load for using the product (again, tap comes with a TON of shit I couldn’t care less about but I don’t really care…)

[yoshuawuyts]

very good point - closing for now!