[Mlir-commits] [mlir] [mlir] Improve dialect conversion failure diagnostics (PR #182729)

[Matthias Springer]

================
@@ -432,7 +432,7 @@ func.func @test_lookup_without_converter() {
 // expected-remark at -1 {{applyPartialConversion failed}}
 
 func.func @test_skip_1to1_pattern(%arg0: f32) {
-  // expected-error at +1 {{failed to legalize operation 'test.type_consumer'}}
+  // expected-error at +1 {{failed to legalize operation 'test.type_consumer' that was explicitly marked illegal: "test.type_consumer"\((%[0-9]+|<<UNKNOWN SSA VALUE>>)\) : \(f32\) -> \(\)}}
----------------
matthias-springer wrote:

I'm not sure if that works. You can try `expected-error-re` or just drop everything after the operation name.

Why is this needed anyway? `%[0-9]+|<<UNKNOWN SSA VALUE>>` Is the output not deterministic?

https://github.com/llvm/llvm-project/pull/182729