OK, rereading that issue, I disagree with Tarek and I think that the patch on that issue is ill-advised as it looks like it changes behavior in a non-backward-compatible way.

If you changed your set_rw onerror handler to a rm_ro_file error handler, would things work?  That is, have the handler delete the file?

Adding a 'retry' capability is interesting, but would be a non-trivial change in logic, and should be addressed in a new issue, not this one.