[python] TokenStreamRewriter _reduceToSingleOperationPerIndex error when reducing mixture of Replace and Insert operations #4818

Description

The Python TokenStreamRewriter class has a bug when there's a mixture of both Replace and Insert operations queued up, and some of the Insert operations overlap. The "Walk inserts" section of _reduceToSingleOperationPerIndex uses the wrong index when combining Inserts, which risks setting the wrong operation to None.

To Reproduce

1. Define a list of rewrites which has a Replace, then two Inserts which overlap. Eg [ReplaceOp(tokens, 2, 4, "foo"), InsertBeforeOp(tokens, 6, "bar"), InsertBeforeOp(tokens, 6, "baz")]
2. Try call _reduceToSingleOperationPerIndex (either directly in a unit test, or by calling getText()
   • Expected behaviour: should combine the Insert operations and leave the ReplaceOp untouched, leaving: [ReplaceOp(tokens, 2, 4, "foo"), None, InsertBeforeOp(tokens, 6, "bazbar")]
   • Actual behaviour: Incorrectly sets the ReplaceOp to None because the Index reference is wrong, leaving: [None, InsertBeforeOp(tokens, 6, "bar"), InsertBeforeOp(tokens, 6, "baz")]
3. It then fails with a ValueError because two different operations are still defined for the same index, 6.

Suggested Fix

The "# Walk Replaces" section of the same method is already solving this issue by referencing .instructionIndex. I suspect that Inserts don't do the same is an oversight. So I'd suggest implementing the same approach for Inserts checks as well:

1. Change both line 177 and line 180:
   • from: rewrites[prev_index] = None
   • to: rewrites[prevIop.instructionIndex] = None
2. After that change, prev_index is no longer needed anywhere, so we can remove the enumerate call and change line 174:
   • from: for prev_index, prevIop in enumerate(prevInserts):
   • to: for prevIop in prevInserts: (this is same code as the Walk Replace section)