[Mlir-commits] [mlir] [mlir][arith] Fix crash in IntRangeOptimizations due to stale solver state (PR #186187)

[Mehdi Amini]

================
@@ -148,3 +148,107 @@ func.func @analysis_crash(%arg0: i32, %arg1: tensor<128xi1>) -> tensor<128xi64>
   %2 = arith.extsi %1 : tensor<128xi32> to tensor<128xi64>
   return %2 : tensor<128xi64>
 }
+
+// -----
+
+// Make sure int-range-optimizations does not crash when potential control-flow folding
----------------
joker-eph wrote:

See my first comment in the PR:

>  The test is very fragile unfortunately, which also comes with the nature of the bug (it only happens when a pointer is reused...). I struggled a bit to have a consistent trigger here.

I'm not sure what to do, not adding a test at all?

https://github.com/llvm/llvm-project/pull/186187