[Mlir-commits] [mlir] [mlir][Func] Preserve attribute when converting CallOp/ReturnOp signature (PR #127772)

[Hongren Zheng]

================
@@ -126,8 +127,9 @@ class ReturnOpTypeConversion : public OpConversionPattern<ReturnOp> {
   LogicalResult
   matchAndRewrite(ReturnOp op, OneToNOpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const final {
-    rewriter.replaceOpWithNewOp<ReturnOp>(op,
-                                          flattenValues(adaptor.getOperands()));
+    rewriter
+        .replaceOpWithNewOp<ReturnOp>(op, flattenValues(adaptor.getOperands()))
+        ->setAttrs(op->getAttrs());
----------------
ZenithalHourlyRate wrote:

The design of these callbacks could be discussed in the discourse and I'm happy to be tagged and involved there.

The decision for this PR is then accepting, rejecting or pending and waiting for the discussion above to materialize? I still prefer a consistent behavior among func::FuncOp/CallOp/ReturnOp for now and the default behavior, as the downstream has no ability to manipulate them now, should be preserving.

For example, there might be some attribute set to the func::FuncOp for some purpose, which I believe is fairly common, and then a simple DialectConversion pass will clear it and the user has to workaround it everywhere. 

https://github.com/llvm/llvm-project/pull/127772