Auto-import of modules plus inability to order completions leads to pathalogical cases

[mqudsi]

The inability to influence the order of completions combined with RA suggesting completions from as-of-yet un-imported traits can result in the completions being rendered virtually useless by the presence of certain dependencies that pollute the RA's suggested completions list and completely overwhelm the actually desirable results (which end up being displayed after or mixed in throughout).

Here's a screenshot showing how having a dependency on color_eyre results in dozens of unwanted completions overwhelming the completions list:

Is there any workaround to suppress the auto-import of a particular crate for the purposes of generating completions (i.e. never suggest anything from color_eyre unless I already have the trait imported)?

[davidbarsky]

There isn't, but I think a configuration option that takes a list of crates to exclude from the flyimport functionality could be a reasonable thing to have.

[mqudsi]

Thanks, that's good to hear. I don't mean to nitpick but I was thinking of blacklisting more on a namespace (module path) rather than crate basis. But I imagine that's what you ultimately envisioned? Then again, perhaps that doesn't work with how RA is currently architectured? Would doing it on a per-crate basis speed up completions overall (because RA doesn't have to walk that crate) or since this is being applied at the auto-import level (rather than lower than that) would it not make a difference? (I am thinking it terms of dual-usage here, if this can be used both to prevent completions pollutions and to speed up RA when there are dependencies it chokes on.)

How doable do you think this is for someone not at all familiar with the codebase to tackle as a PR? Because it sounds somewhat straightforward enough to me, but I've misled myself many a time before!

[Veykril]

If we blacklist per crate that would be a speed improvement as we can just skip searching for paths for items from blacklisted crates (but this speed up is strictly for completions). Blacklisting module segments instead would still be a speed up but significantly less so as we would merely be skipping looking at paths from blacklisted modules (that is we keep looking on alternative paths where as the other approach doesn't even try to find anything). The crate one is certainly simpler to implement and probably what I'd prefer.

rust-analyzer/crates/ide-completion/src/completions/flyimport.rs

Line 266 in 9fd6c69

.search_for_imports(&ctx.sema, import_cfg, ctx.config.insert_use.prefix_kind)

is the entry point for this in import completions. You need to drill through that a bit to get to the part that calculates applicable items, which happens here

rust-analyzer/crates/ide-db/src/items_locator.rs

Line 17 in 9fd6c69

pub fn items_with_name<'a>(

. So the callers of that function should filter out items from blacklisted crates in some manner

[Veykril]

It looks straight forward to me, just a bit annoying as the required change is several layers deep and you need to thread the config through there

[lnicola]

We could also offer just one completion, like yellow (+100 others) when the trait is not in scope.

[BenjaminBrienen]

Is there a way to make folders of sorts? Group them by crate, module, or trait and select the group to expand the subcontext menu.

[lnicola]

I don't think the LSP supports anything like that.

[BenjaminBrienen]

I thought not :(

[ChayimFriedman2]

Turns out I'm working on this without intending :P and already close to an end.

So
@rustbot claim