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.
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.
In this commit tests were added to ensure that tests with a `const` index behaved as expected.
In order to minimize the changes to the test's corresponding `stderr`, the tests were appended to
the end of the file.
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.
