[Mlir-commits] [mlir] [mlir][sparse] Fix crash in sparsification when disjunctive op has sparse operand (PR #184599)

[Aart Bik]

================
@@ -1402,7 +1402,6 @@ Merger::buildTensorExp(linalg::GenericOp op, Value v) {
     // is sparse. For a disjunctive operation, it yields a "sparse" result if
     // all operands are sparse.
     bool conjSpVals = xSpVals || ySpVals;
-    bool disjSpVals = xSpVals && ySpVals;
----------------
aartbik wrote:

> I'm absolutely not confident about it, and I was surprised no test failed. I was expecting one to provide a counter example :)

I was thinking the same thing and then realizing we may have done a poor job testing this ;-(

https://github.com/llvm/llvm-project/pull/184599