zygoloid wrote:
> > Another possibility to consider for the original bug: `(single_identifier)`
> > is almost certainly a cast, not redundant parentheses, unless
> > `single_identifier` names a macro argument. So I wonder if that would be a
> > better heuristic to use to fix the regression.
>
> I don’t think that would work as`(identifier)` can be a number of constructs
> depending on the context. It can be part of a function/macro call, function
> declaration, macro definition, C++ cast, C cast, `sizeof` operand, etc.
OK, but the context is deciding whether `(identifier)&x` or `(identifier)*x` is
a binary operator or a cast. Given no other information, the fact that it's a
single identifier seems like a very strong signal that it's a cast.
I don't think this PR fixes #83400 -- handling a small handful of common types
won't address the same issue for the much larger class of cases where the type
name is not one of these types. Eg, in #83400 itself the examples included:
```diff
- static _FORCE_INLINE_ m_type get(const Variant *v) { return
(m_type)*VariantInternal::get_int(v); } \
+ static _FORCE_INLINE_ m_type get(const Variant *v) { return
(m_type) * VariantInternal::get_int(v); } \
```
But looking for a parenthesized single identifier addresses all of the examples
in the issue.
https://github.com/llvm/llvm-project/pull/83709
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
