[Mlir-commits] [flang] [mlir] Make MLIR Value more consistent in terms of `const` "correctness" (NFC) (PR #72765)

[Mehdi Amini]

================
@@ -335,8 +335,9 @@ namespace llvm {
     MutableArrayRef(T *begin, T *end) : ArrayRef<T>(begin, end) {}
 
     /// Construct a MutableArrayRef from a SmallVector.
-    /*implicit*/ MutableArrayRef(SmallVectorImpl<T> &Vec)
-    : ArrayRef<T>(Vec) {}
+    template <typename U>
+    /*implicit*/ MutableArrayRef(const SmallVectorTemplateCommon<T, U> &Vec)
+        : ArrayRef<T>(Vec) {}
----------------
joker-eph wrote:

>  Not sure I follow that - most changes allow things to compile that didn't before (adding a new function, etc), so I'm not sure I follow that being a relevant criteria. 

I feel you're being pedantic about it here: there is no possibe observable behavior different in the compiler behavior, there is code coverage in the project: hence that seems "good enough" for me.
Of course can go the extreme way in terms of testing, but I repeat that this isn't in line with the way we're testing in general: even if ADT has more unit-tests by nature, the "100% coverage through C++ unit-tests" isn't I believe a generalized trade-off we make in LLVM.
Most "NFC" patch are shuffling code around in a way that makes something compile/work that wasn't before, and yet we don't write unit-tests for every single change.

https://github.com/llvm/llvm-project/pull/72765