Changes like this one aren't in main because #57772 added checks/transforms to prevent them. Here, for example, now we lookup B successfully and think "hey, I'll just reuse the node" (which also happens with the isDeclarationName fix added in #57772, for this specific case), but the node's got a modifier that needs deleting and a : any that needs adding that it doesn't add, while in #57772 I added a line to the visitor to add the missing : any and drop the public from existing nodes in the node builder. There's a couple changes like this in here. They're not that bad a change because it's not like this node's form nowhere - it's the input, so it's at least no worse than the input - but seeing an error in the input (like this) and just dumping the erroneous node into the output is something we usually try to avoid. We could also backport the changes to visitExistingNodeTreeSymbols from that PR if we want to minimize the diff.

Does that mean that this PR is wrong without that other PR? Or is this PR safe to include in a patch by itself?