[incremental] remove unnecessary passes

[nikomatsakis]

This is a sub-issue of #40746. The follow list of tasks are all executed for side-effects. I believe they can all be straight-up removed (that is, we just don't create a task at all). The idea is to remove the call to with_task or in_task as well as the associated DepNode variant and see what happens -- if you get assertion failures, it may mean that they are writing to shared state, in which case we'll want to keep the task and refactor it in some other way.

An example PR is #41063. See e.g. the first commit which removed the CheckLoops stuff.

• src/librustc_passes/loops.rs:53: let _task = map.dep_graph.in_task(DepNode::CheckLoops);
• src/librustc_passes/static_recursion.rs:91: let _task = hir_map.dep_graph.in_task(DepNode::CheckStaticRecursion);
• src/librustc_borrowck/borrowck/mod.rs:64: tcx.dep_graph.with_task(DepNode::BorrowCheckKrate, tcx, (), check_crate_task);
• src/librustc/middle/effect.rs:244: let _task = tcx.dep_graph.in_task(DepNode::EffectCheck);
• src/librustc/middle/liveness.rs:199: let _task = tcx.dep_graph.in_task(DepNode::Liveness);
• src/librustc_typeck/check_unused.rs:65: let _task = tcx.dep_graph.in_task(DepNode::UnusedTraitCheck);
• src/librustc/middle/dead.rs:597: let _task = tcx.dep_graph.in_task(DepNode::DeadCheck);
  • this would use the AccessLevels query described elsewhere
• src/librustc_typeck/coherence/mod.rs:135: let _task = tcx.dep_graph.in_task(DepNode::Coherence);
• src/librustc_typeck/coherence/overlap.rs:42: tcx.dep_graph.in_task(DepNode::CoherenceOverlapCheck(trait_def_id));
• src/librustc_typeck/collect.rs:171: self.tcx.dep_graph.with_task(DepNode::CollectItemSig(def_id), self.tcx, id, op);

[nikomatsakis]

@aochagavia expressed interest in this from #40746

[aochagavia]

By the way, in the issue you mention that src/librustc/middle/dead.rs:597 needs to use an AccessLevels query. However, it seems that it is already being used on line 598: let access_levels = &ty::queries::privacy_access_levels::get(tcx, DUMMY_SP, LOCAL_CRATE);

[aochagavia]

@nikomatsakis I have a commit where I remove 4 of the 5 passes listed here. However, now I am blocked by the following question:

The pass in src/librustc_typeck/coherence/mod.rs:135 used the DepNode::Coherence variant. After deleting the pass, I wanted to delete DepNode::Coherence as well, but I discovered that it is being used in multiple places. Particularly, it is being used by this function, which in turn is being used for something related to maps in rustc/ty/maps.rs (check out this line). Should those maps also be removed? I am unable to see if they are somehow related to the task I just deleted. Maybe I should just let DepNode::Coherence alone.

[aochagavia]

Also, the pass in src/librustc_typeck/collect.rs:171 is part of a function that clearly has side effects. In fact, removing the pass would leave the function body empty. What course of action would you suggest here?

[eddyb]

Yeah, don't remove nodes used by maps. Those are on-demand queries.

[aochagavia]

@eddyb do you have any ideas regarding my second question?

[eddyb]

@aochagavia I think you'd want to call op instead of removing it altogether, i.e.:

        self.with_collect_item_sig(item.id, convert_item);
// becomes
        convert_item(self.tcx, item.id);

[nikomatsakis]

@aochagavia argh, just saw your comments here. All resolved?

I've been behind on notifications. I wound up accidentally doing some of this work in #41360, actually. Sorry if I stepped on your toes with that. :(

[aochagavia]

@nikomatsakis There is indeed some overlap, so I will have to rebase once your changes land. On the other hand, if changes need to happen fast here maybe I could work on other parts of the compiler (for instance, this issue)

[Mark-Simulacrum]

@nikomatsakis All the items in the issue are marked off, does that indicate this is done? If not, maybe OP should be updated with more items or a note added that there are more issues as well.

[nikomatsakis]

I think this is done -- @aochagavia, you agree?

[aochagavia]

@nikomatsakis I don't understand the question. What should I agree on? If this is done, it is done, right?

[nikomatsakis]

@aochagavia true. =) I'm not sure precisely what I was asking for. (I guess if you had noticed anything I overlooked, but if so, we'll find it sooner or later...)