liaison: 2026-05-27 expert session with t-clippy (#677)

Author: PLeVasseur

subcommittee/liaison/meetings/2026-05-27-clippy-expert-session/minutes.md

@@ -0,0 +1,199 @@
+# Safety-Critical Rust Consortium Meeting on 2026-05-27 @ 10am EDT / 4pm CEST
+
+Expert session with Jason Newcomb of the Rust Project Clippy Team (t-clippy)
+
+| Search Key | Description |
+| :---- | :---- |
+| \[todo\] | Action Item |
+| \[decision\] | Something decided on |
+| \[important\] | Key information |
+
+## Agenda
+
+1. Introduction to t-clippy  
+2. Practical guide to writing Clippy lints
+
+## Check-in area
+
+**Please add your name, and an emoji that describes your day.**
+
+* David Svoboda (-:  
+* Jason Newcomb  
+* Michael Henn 🥵  
+* Manuel Hatzl ☀️  
+* Samuel Wright 🌞  
+* Arshad Mahmood 🌅  
+* Max Jacinto 🍀  
+* Pete LeVasseur 🦀🦺  
+* Mira Baumann 🧵  
+* Raiki Tamura🏫  
+* Achim Kriso 🦆  
+* Mikhail Antoshkin 💤  
+* William Barsse  
+* Marcos Borges 😎  
+* Christof Petig 🥵
+
+**Notetaker:**
+
+* Pete LeVasseur
+
+## Housekeeping section
+
+* 
+
+## Tasks
+
+* 
+
+## Meeting Minutes
+
+## Introduction to t-clippy
+
+Clippy Book / Getting Started Guide
+
+* [https://doc.rust-lang.org/nightly/clippy/development/infrastructure/book.html](https://doc.rust-lang.org/nightly/clippy/development/infrastructure/book.html)
+
+Who has worked on/with Clippy?
+
+* Pete LeVasseur  
+* Arshad Mahmood  
+* Achim Kriso
+
+Who has worked on/with static analysis tools in the past?
+
+* David Svoboda
+
+Differing levels of where you can “hook into”
+
+* Syntax/AST  
+* HIR  
+* MIR
+
+Can we have the lints be put into their own Safety-Critical Rust grouping (to Jason Newcomb)?
+
+* Jason will bring this forward to the team and discuss the idea  
+* Might or might not be a need for this depending on (1) how many lints we are writing (2) how 
+
+Can we have lints that overlap between different groups?
+
+* Technically possible today  
+* Prefer to not do this  
+* Jason would need to bring this back to the team to discuss the usecase of safety-critical to be able to “turn on all the things”
+
+Starting point
+
+* `cargo dev new_lint -h` for reading how to create a new lint  
+* Also read: [https://doc.rust-lang.org/nightly/clippy/development/infrastructure/book.html](https://doc.rust-lang.org/nightly/clippy/development/infrastructure/book.html)  
+* Basic scaffolding will be set up  
+  * “Lint pass” to hook into compiler’s infrastructure  
+    * Could be any number of lints in a single lint pass  
+* Lint pass runs as visitor over entire HIR tree  
+  * Can then define `check_expr` on the `LateLintPass` which will run over every expression node in the HIR tree  
+  * Can then define `check_body` as well as an alternative type of item to check  
+  * Many, many hooks are available to you in here to hook into the visitor
+
+In `check_expr`:
+
+* Tend to first do some sort of filtering to find the type of expression that you’d like to lint on  
+  * `match e.kind { /* .. */ }`  
+  * Can then filter onto e.g. `ExprKind::Path`  
+    * Can allow for destructuring and further filtering  
+* Entails reading a lot of compiler documentation to understand how to check over HIR  
+  * [https://doc.rust-lang.org/nightly/nightly-rustc/](https://doc.rust-lang.org/nightly/nightly-rustc/)  
+  * [https://rustc-dev-guide.rust-lang.org/](https://rustc-dev-guide.rust-lang.org/)  
+  * [https://github.com/rust-lang/rust-clippy/tree/master/book/src/development](https://github.com/rust-lang/rust-clippy/tree/master/book/src/development)  
+  * Some things are documented, some things are not  
+  * May, at times, need to look at compiler source code  
+* Can further filter, e.g.  
+  * `match p { QPath::Resolved(/* .. */) => { /* .. */ }, /* .. */ }`  
+* Many, many hooks are available to you in here
+
+The tests and how to work with them
+
+* Ends up being named the same as the lint that you named  
+* All lints run on all files (dogfooding)  
+* If a new lint triggers on other files, you can tell them to ignore the new lint  
+  * When writing test, then default enabled things run at the same time  
+  * Could cause it to take some time to run at first  
+  * Normally then handle this with a targeted `#![allow(clippy::other_failing_lint_we_dont_care_about)]`  
+* There are `your_lint_here.stderr` files which contain the expected output of the lint that this is checked against
+
+The mindset
+
+* Write a test which is “normal” and represents where the lint should fire  
+* Write the the lint logic  
+* Go run the lints  
+* Note that it may fail or not catch what we want  
+* Once it does, run the `bless` command and go back and write another test, then flesh out the lint more in a loop
+
+Question: Is there some kind of workflow / secret that will not be on the documentation?
+
+* No, not really  
+* All of what Jason is walking us through is available in the documentation  
+* Some of the things “not in” the Clippy documentation could be where as a lint implementor you have to get into how the compiler works, ASTs, HIR, MIR
+
+Question: How about a more complex lint? Like constant propagation?
+
+* Please hang tight, we’ll come back to this
+
+`LateLintPass` details for HIR
+
+* Doesn’t contain type information  
+* Second parameter of `LateContext` will give useful information  
+  * `cx.typeck_results()` will contain all the type checking information  
+  * Can work with this further, e.g. `typeck.expr_ty(e)`  
+* There is a helper function to use for checking on trait implementation  
+  * `implements_trait!()`  
+* Also have access to e.g. `cx.tcx.lang_items().eq_trait()` for getting the trait’s definition ID  
+* Depending on if the trait has generics, there’s now an assert which forces passing this information that’s relevant in  
+* Can resolve method calls, e.g. `ExprKind::MethodCall(/* .. */) => { def_id = typeck.type_dependent_def_id(e.hir_id) }`  
+  * This information must be obtained like above  
+* Other useful things like diagnostic items, e.g. `cx.tcs.is_diagnostic_item(name, did)`  
+  * You’ll need to check the compiler documentation for this  
+  * Jump to definition will take you to compiler source so you can get that info that’s decorated on the item: `rustc_diagnostic_item`  
+* Helpful at times when filtering to “do the opposite of what the check is for”  
+* When referring to typechecked item, goes back to the trait, not the implementation of the trait  
+  * If “is this String’s implementation of something?” have to check type and Definition ID (def\_id)
+
+`LateLintPass` details for MIR
+
+* `check_fn()` or `check_body()`  
+* `check_fn()` may also handle closures  
+* First need to write a check for if MIR can be obtained  
+  * If something can never be satisfied, then the MIR body never exists  
+  * Would crash if trying to obtain  
+* Work on optimized MIR  
+* Perform whole MIR lint check on the body  
+* For data flow analysis there’s fixed point analysis framework that allows to go MIR block by MIR block using the `Analysis` trait ([rustc\_mir\_dataflow](https://doc.rust-lang.org/beta/nightly-rustc/rustc_mir_dataflow/))  
+  * Propagate until you run out of checks/changes to perform  
+* Not a lot specific to doing const propagation  
+  * Likely need to write your own code to handle this
+
+Because lints are run on macro-expanded code
+
+* You have to do some things to figure out if that’s the case  
+* On each node you have the span  
+  * `e.span.ctxt()` \=\> syntax context  
+* Unique syntax context for each  
+  * macro expansion and   
+  * AST \=\> HIR lowering  
+* Syntax context allows for telling when this expansion or lowering has occurred  
+  * `e.span.ctxt().outer_expn_data()`  
+  * Would end up with two different macro expansion context  
+* When writing lints you may need to keep track of this; these can show up anywhere  
+  * Any given `x` and `y` might originate from a macro expansion  
+  * For style lints this matters and you have to compare context a lot along the way  
+  * `e.span.ctxt() == p.span.ctxt()`  
+    * Checking if the context of the parent thing is the same as the expression under investigation
+
+Ways to solicit help / support:
+
+* If writing a lint and stuck \=\> open a draft PR  
+* If stuck, can open a topic on the Rust Zulip’s clippy channel
+
+## Material
+
+Any material to read before the meeting should be included here.
+
+* 
+