feat: Add profile-specific configuration for disallowed methods and types#15779
Conversation
|
@blyxyas ping |
|
Hi @Lallapallooza, |
There was a problem hiding this comment.
I'll recount what I understand here, so a user uses a special configuration for settings profiles in their clippy.toml so any function can opt-in to those profiles, enabling warnings to those designated disallows in that function.
For example, one might want to disallow functions that may execute arbitrary code from a string input, into user-input-handling functions, and this is a way to achieve via Clippy warning of those uses.
Am I correct, have I forgotten/misunderstood something? If that's the case, we might be interested in generalizing this a bit so that both (1.) we can use profiles for other lints without the need of adding a special attribute for each lint, and (2.) people can use the general idea of profiles in Clippy to segment their source code for several lints at the same time.
Also, @ojeda could this be interesting for the Rust4Linux project? Should I give it priority?
|
You read it correctly. This PR introduces profile-scoped disallow lists in I agree that generalizing this is a good idea. I can follow up with a patch if that would be useful. Do you have any preferences for the design or the user interface (i.e., how Clippy users would enable and use profiles)? |
|
Thanks @blyxyas for the ping! It is an interesting idea. IIUC, currently this can only be done at a crate granularity, right? So it is essentially a way to make Clippy more granular -- I agree that generalizing it makes sense. I will try to think if we could use it in Rust for Linux (@Lallapallooza What are the use cases that motivated this? It may help to give some in the docs to inspire others -- thanks!). |
|
Hi @ojeda. My initial contribution was motivated by a need in my project to disallow host-side operations (e.g., any conversion from a device tensor to a host tensor) in specific context (e.g. dir crates/lib/cuda/) so that such code fails to compile. |
I'll ask both the entire team, and the style team to see if they have anything to say. In Clippy we don't really have an in-house attribute style guide, so maybe they have some feedback about how this could be implemented. |
|
Thanks @Lallapallooza, that is good to know (if you have more cases you can think of, then it would be great to hear about them, of course). |
This comment has been minimized.
This comment has been minimized.
|
Hi Lallapallooza, I've talked with the team and I'm going to proceed with the review. Also, sorry for the delayed response, I've just returned from vacation. I was thinking that this may be useful for this message thread
I thought that we already had a configuration option for this, but it seems that we don't have one. So I think that this feature can also support this use case. |
a8c3a91 to
0ff61e7
Compare
This comment has been minimized.
This comment has been minimized.
|
@blyxyas Thanks, I have rebased pr. |
I'm walking back this statement, we already have another pull request taking care of this. So we won't focus on that front here. |
|
Reminder, once the PR becomes ready for a review, use |
|
Hi! I have addressed all comments, could you please check again. |
| pub fn active_profiles(&mut self, cx: &LateContext<'_>, hir_id: HirId) -> Option<&ProfileSelection> { | ||
| if self.cache.contains_key(&hir_id) { | ||
| return self.cache.get(&hir_id).and_then(|selection| selection.as_ref()); | ||
| } |
There was a problem hiding this comment.
self.cache.contains_key is only self.cache.get().is_some(), so it's performing the same lookup two times. It should be something more akin to:
if let Some(selection) = self.cache.get(&hir_id) {
return selection;
}
There was a problem hiding this comment.
added a NOTE why it's needed
|
Hi @blyxyas, sorry for leaving this PR sitting for so long. I've rebased onto current master and addressed all the outstanding review comments (docs on Replies are in each thread. Whenever you have a moment, could you take another look? Thanks again for the reviews. 🙏 |
|
@blyxyas Hi, could you check my patch again please. |
This comment has been minimized.
This comment has been minimized.
- document `ProfileEntry` and `ProfileSelection` - explain why `profiles` and `known_profiles` are separate - emit unknown-profile warning inline (drop intermediate SmallVec) - add tests covering `#[expect]` before/after `#[clippy::disallowed_profile]` with an unknown profile name, for both methods and types lints - sort_by -> sort_by_key to satisfy dogfood
647bc06 to
03128d7
Compare
|
This PR was rebased onto a different master commit. Here's a range-diff highlighting what actually changed. Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers. |
|
@blyxyas Thanks! I rebased MR and now it's mergeable. |
|
Why did this not get an FCP? I remember there being a zulip thread quite a while ago that had some objections that got left hanging. |
|
I'll revert this pull request, I forgot about those objections (I'll look them up right now). I'm sorry for all nightly users that got an un-FCP'd feature in their nightly. |
|
The objection was more a vague vibes based objection, not any specific issues. Not an insurmountable thing in any way, but it should've been addressed before this was merged. |
|
This has now been reverted. Again, this isn't rejection of the feature in some form (or even the current form). The issue here is that this is a notable config change/lint extension which had some concerns raised which weren't addressed before it was merged. |
Thanks. Please let me know after the FCP if any changes are required from me. I’m happy to adjust the patch based on the requirements. |
…ds and types" (rust-lang#17251) Reverts rust-lang#15779 due to un-FCP'd concerns. This is wholy my fault. I didn't want to leave a user hanging and didn't do my due-diligence of getting all FCP concerns checked out and resolved. r? @Jarcho cc @Lallapallooza changelog: [`disallowed_methods`]: Revert profile-specific configuration changelog: [`disallowed_types`]: Revert profile-specific configuration
|
Can you open a new PR please. It makes tracking things easier if the FCP is linked to an open PR. |
|
View all comments
Add profile-scoped disallow lists for methods and types, wiring the new configuration tables through a shared resolver that can be toggled with #[clippy::disallowed_profile] attributes.
changelog: [disallowed_methods]: allow selecting per-scope disallow lists via disallowed-methods-profiles and the clippy::disallowed_profile attribute
changelog: [disallowed_types]: allow selecting per-scope disallow lists via disallowed-types-profiles and the clippy::disallowed_profile attribute