RFC - Dynamic source conditions

[levouh]

Adds a dynamic_condition that is evaluated at "runtime" for all sources that discerns whether or not it will be run each time they have the potential to be run.

See associated discussion here as to why this might be useful.

Some things still to do, but curious on your thoughts @jose-elias-alvarez before messing with getting these going:

• Add tests

[jose-elias-alvarez]

This is a great idea and works well. Strictly speaking, since null-ls allows running generators on demand, something like this is already possible, but I think this is a clean addition and far more user-friendly.

I wonder if it might be worth it to move the dynamic_condition check to the generator loop in generators.run. This would give the callback access to params, which I imagine might be useful to simplify checking whether a source should run for the given file or not. Maybe something like this:

diff --git a/lua/null-ls/generators.lua b/lua/null-ls/generators.lua
index 41510af..9aea507 100644
--- a/lua/null-ls/generators.lua
+++ b/lua/null-ls/generators.lua
@@ -18,6 +18,12 @@ M.run = function(generators, params, postprocess, callback)
         for _, generator in ipairs(generators) do
             table.insert(futures, function()
                 local copied_params = vim.deepcopy(params)
+
+                local dynamic_condition = generator.opts.dynamic_condition
+                if dynamic_condition and not dynamic_condition(copied_params) then
+                    return
+                end
+
                 local to_run = generator.async and a.wrap(generator.fn, 2) or generator.fn
                 local ok, results = pcall(to_run, copied_params)
                 a.util.scheduler()
@@ -92,19 +98,9 @@ M.run_registered_sequentially = function(opts)
     M.run_sequentially(generators, make_params, postprocess, callback, after_all)
 end

What do you think? Other than that, I'm happy to merge this once we have tests (unit tests and a simple E2E test would be ideal).

[levouh]

Makes sense to me, having access to params would be nice as well. I also think a name like runtime_condition might be better as that's what we're doing, but curious on your thoughts there as well.

I'll look at this patch and adding tests to get it through sometime this week. Thanks for the feedback.

[levouh]

Let me know how that looks @jose-elias-alvarez.

[jose-elias-alvarez]

Looks great, thank you for working on this (and for the improvements to the documentation)!