| Issue |
86714
|
| Summary |
[clang-tidy] Create check for inconsistent/missed typedef/using alias type names
|
| Labels |
check-request
|
| Assignees |
|
| Reporter |
FruitClover
|
Consider the following example: https://godbolt.org/z/jE8Yc7sz6
```c++
using Type = int;
struct Base {
virtual Type fooType(Type x);
virtual int fooInt(int x);
};
struct Derived : Base {
int fooType(int x);
Type fooInt(Type x);
};
```
Here we misuse/miss consistent **alias names** in derived methods (compared to base
class signatures), which could lead to tricky bugs (after `using Type = ...`
changes) and this is also complicates plain text search.
The idea is that we could detect such **`using`/`typedef`** inconsistency/misuse in
some cases with clang-tidy.
To start up, we could focus on:
- Class methods only.
- Return type check.
- Argument type check.
To illustrate which cases are compatible/which are not - check the following table:
<details>
<summary>legend</summary>
- `Base Type`: return/argument type for class class methods.
- `Derived Type`: return/argument type for derived class methods.
- `Warning/FixItHint`: if check should emit warning/hint
(we use `int` as some non-aliased type, and <alias> as `using T = ..` or
`typedef .. T`;
</details>
```
| | Base Type | Derived Type | Warning | FixItHint |
|-----+-----------+--------------+---------+-----------|
| | int | int | - | - |
| | <alias> | int | + | + |
| | int | <alias> | + | - |
| (*) | <alias> | <alias> | Analyze | - |
```
`(*)` **Both types are aliased**: Analyze `BaseT` vs `DerivedT`:
1. Alias names are compatible: (**Warning: No**)
1.1 one of (`BaseT`, `DerivedT`) is implicit
2.1 Same fully-qualified names:`BaseT.getQualifiedNameAsString == DerivedT.getQualifiedNameAsString`
2. Alias names are non-compatible: (**Warning: Yes**)
2.1. Strict check: `TD1.getQualifiedNameAsString != TD2.getQualifiedNameAsString` (`N1::Type vs N2::Type`)
3. Questinable: (**Warning: ?**)
3.1. chain alias (maybe control with option). If so, should check it before 2.1
```c++
using T1 = int;
using T2 = T1;
// => T1 name is compatible with T2 name
```
This probably belongs to `bugprone` category, something like `inconsistant-alias-names`/`inconsistant-using-typedef` (suggestions are welcome)
Background: PoC version was prepared for internal projects, here I want to check if clang-tidy would be interested in it before publishing PR, and discuss what to warn and if we should emit more fix hints.
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
