This is a follow-up to #2951 that extends the logic to allow for
returning references inside structs/enums/unions. This was a simple
oversight in the first version and it's surprisingly easy to handle.
The `needless_borrow` lint is temporarily disabled because of some false
positives it causes in combination with the `derive` macro.
However the documentation does not explain these issues, but instead
lists `Known problems: None`. To make it clear why this lint is
currently not enabled, a description of the false positives caused by
this lint has been added to the `Known problems` section.
Currently this code will trigger `trivally_copy_pass_by_ref`:
```
struct OuterStruct {
field: [u8; 8],
}
fn return_inner(outer: &OuterStruct) -> &[u8] {
&outer.field
}
```
If we change the `outer` to be pass-by-value it will not live long
enough for us to return the reference. The above example is trivial but
I've hit this in real code that either returns a reference to either the
argument or in to `self`.
This suppresses the `trivally_copy_pass_by_ref` lint if we return a
reference and it has the same lifetime as the argument. This will likely
miss complex cases with multiple lifetimes bounded by each other but it
should cover the majority of cases with little effort.
The macro always negates the result of the given comparison in its
internal check which automatically triggered the lint. As its an
external macro there was no chance to do anything about it which lead
to a white listing of all external macros to prevent further issues.
This commit removes the logic in this PR that linted out-of-bounds constant `usize` indexing on arrays. That case is already handled by rustc's `const_err` lint. Beyond removing the linting logic, the test file and its associated stderr were updated to verify that const `usize` indexing operations on arrays are no longer handled by this `indexing_slicing` lint.
This commit contains a few changes. In an attempt to clarify which test cases should and should not produce stderr it became clear that some cases were being handled incorrectly. In order to address these test cases, a minor re-factor was made to the linting logic itself.
The re-factor was driven by edge case handling including a need for additional match conditions for `ExprCall` (`&x[0..=4]`) and `ExprBinary` (`x[1 << 3]`). Rather than attempt to account for each potential `Expr*` the code was re-factored into simply "if ranged index" and an "otherwise" conditions.
The changes reflected in this commit (requested in PR #2790) are as follows:
- Extended `INDEXING_SLICING` documentation to include the array type so that it is clearer when indexing operations are allowed.
- Variable `ty` defined identically in multiple scopes was moved to an outer scope so it's only defined once.
- Added a missing return statement to ensure only one lint is triggered by a scenario.
- Prettified match statement with a `let` clause. (I learned something new!)
- Added `&x[5..].iter().map(|x| 2 * x).collect::<Vec<i32>>()` and `&x[2..].iter().map(|x| 2 * x).collect::<Vec<i32>>()` to the test cases. The first _should trigger the lint/stderr_ and the second _should not_.
The changes reflected in this commit are as follows:
- Revised `IndexingSlicingPass` struct name to IndexingSlicing for consistency with the rest of the code base.
- Revised match arm condition to use `(..)` shorthand in favor of `(_, _, _)`.
- Restored a couple telling variable names.
- Calls to `cx.span_lint` were revised to use `utils::span_help_and_lint`.
- Took a stab at refactoring some generalizable calls to `utils::span_help_and_lint` to minimize duplicate code.
- Revised INDEXING_SLICING declaration to pedantic rather than restriction.
- Added `&x[0..].get(..3)` to the test cases.
This commit renames instances of `array_indexing` to `indexing_slicing` and moves the `indexing_slicing` lint to the `clippy_pedantic` group. The justification for this commit's changes are detailed in the previous commit's message.
Hey there clippy team! I've made some assumptions in this PR and I'm not at all certain they'll look like the right approach to you. I'm looking forward to any feedback or revision requests you have, thanks!
Prior to this commit the `indexing_slicing` lint was limited to indexing/slicing operations on arrays. This meant that the scope of a really useful lint didn't include vectors. In order to include vectors in the `indexing_slicing` lint a few steps were taken.
The `array_indexing.rs` source file in `clippy_lints` was renamed to `indexing_slicing.rs` to more accurately reflect the lint's new scope. The `OUT_OF_BOUNDS_INDEXING` lint persists through these changes so if we can know that a constant index or slice on an array is in bounds no lint is triggered.
The `array_indexing` tests in the `tests/ui` directory were also extended and moved to `indexing_slicing.rs` and `indexing_slicing.stderr`.
The `indexing_slicing` lint was moved to the `clippy_pedantic` lint group.
A specific "Consider using" string was added to each of the `indexing_slicing` lint reports.
At least one of the test scenarios might look peculiar and I'll leave it up to y'all to decide if it's palatable. It's the result of indexing the array `x` after `let x = [1, 2, 3, 4];`
```
error: slicing may panic. Consider using `.get(..n)`or `.get_mut(..n)`instead
--> $DIR/indexing_slicing.rs:23:6
|
23 | &x[0..][..3];
| ^^^^^^^^^^^
```
The error string reports only on the second half's range-to, because the range-from is in bounds!
Again, thanks for taking a look.
Closes#2536
