[Mlir-commits] [mlir] [mlir][dataflow] Fix for integer range analysis propagation bug (PR #93199)

[Spenser Bauman]

sabauma wrote:

> Re that check, I _think_ the point was to prevent having to create fake constant ranges for non-integer inputs.

@krzysz00, yes That check was added explicitly to prevent a crash relating to faking the integer ranges. That problem seems to no longer occur.

> Maybe we should think about this check? Why don't we do setAllToEntryStates() on the Operation's results when any of the operands are unitialized?

@ubfx I agree my proposed solution is kind of odd. As @krzysz00 mentioned in his comment, I suspect removing that check is the right approach, but it would require changes to the `InferIntRangeInterface`. I would be happy to take a stab at that, but it would be a more invasive change.


https://github.com/llvm/llvm-project/pull/93199