[Pre-RFC] - pass manager

[TzviPM]

Initially, this would be useful for the linter, although likely for other tools like the minifier too.

The pass manager would basically allow rules to depend on other rules as well as to register multiple dependent passes. The passes would be stored in a computation DAG and then broken into layers which would end up being the physical passes that are done. For example:

1 - semantics pass 1 (including CFG generation)
2 - semantics pass 2 (additional analysis that requires the CFG to be built)

3 - no_this_before_super lint - pass 1 (note CFG blocks for nodes containing local uses of this before super and which blocks contain super. store this in the rule struct)
4 - no_this_before_super lint - pass 2 for constructor function nodes, walk their CFG to find out which local violations are actual violations or if super was called in a previous block.

Pass DAG:

graph TD;
    2-->1;
    3-->1;
    4-->3;

Loading

This DAG would be broken into 3 layers:

1. semantics pass 1
2. semantics pass 2 with no_this_before_super pass 1 simultaneously
3. no_this_before_super pass 2

I'm curious what others think of this idea, and if it seems desirable, I'll start a more formal RFC.

[bmeck]

I'm interested in this, but would need more concrete usage intent to understand the overall design in terms of things like how this could be used across the tools.

[Boshen]

We'll probably need more use cases for such architectural changes.

As for no_this_before_super, we can hack it by doing a run_once with data saved in its struct (with a RefCell / Mutex).

[Boshen]

Closing as not planned.