Deprecate instant/flags.vim

[dbarnett]

Maktaba's instant/ mechanism for executing files immediately has a few drawbacks:

• Costs overhead on early startup
• Doesn't interact well with various plugin managers
• Encourages overuse

It's especially significant if most plugins have an instant/flags.vim file.

We should find another convention for defining flags that doesn't depend on loading them instantly. It could work something like:

• Flags are defined in whatever standard location for each plugin
• Maktaba queues up Flag.Set operations until the flag is initialized or read.
• Flag.Get and any other attempt to access a flag's value will trigger flag initialization.
• At plugin/ time, all remaining flags are initialized so callbacks and errors trigger.

There'd be some migration pain, and it may not be worth making changes like if there's no measurable performance benefit.

[dbarnett]

Options for where we'd define flags:

• plugin/flags.vim (with the understanding that maktaba might manually source it)
• flags.vim (in the plugin root), always sourced via maktaba
• addon-info.json (but would need a way to accommodate callbacks)

Note they'd still potentially be expensive if they live in plugin/ where vim would try to detect and source them.

Also note: vimdoc would need to be updated to find flags wherever they end up being defined.

[dbarnett]

Dug into details and I'm liking plugin/flags.vim. Then any attempt to Set the flag before the flags file is loaded can just queue up the operation and any attempt to Get a flag can ensure flags are loaded and fully resolved first. vim will source plugin/flags.vim in the common case. It'll need a little special handling to ensure it still throws appropriate errors in corner cases (e.g. configuring nonexistent flag names, plugin doesn't have a plugin/flags.vim file that does maktaba#plugin#Enter).

The PLUGIN.flags dict will need some retooling to avoid being able to access it before it's resolved. Probably that public flags dict should be deprecated and replaced with a PLUGIN.Flags helper. That'll power Glaive's autocomplete stuff so it has correct handling, too.

Finally, we'll want to move towards deprecating instant/ support overall and all the overhead that entails. I'm planning to create a temporary maktaba#EnableLegacyInstantFilesSupport() helper. During some transition period, if users have trouble with some maktaba plugin they can use that override to troubleshoot and get a bug filed with plugin maintainers, but that way you only pay the legacy overhead if you're wanting to run legacy plugins that need it.

[dbarnett]

As a step one we could drop support for other instant files, since flags are going to be a little trickier than the rest. Filed #243 for the general case.

[dbarnett]

I got a proof of concept in #244 except without the feature of trying to queue up flags operations. In the process I realized that in practice the extra cleverness won't save much overhead, because plugins will tend to look up the magic flags like plugin[commands] before vim gets around to loading the plugin/flags.vim file anyway.