Is it really a problem if the old value is deallocated? It sounds like a similar case to <https://bugs.python.org/issue26168#msg259242>, and would only be a problem if you passed a borrowed reference, and relied on the reference staying alive for another argument.

I do like the separate do_ignore() function in patch 4, but I don’t think it is worthwhile allocating a temporary tuple to save values in. The allocation can fail. Also, I understand do_mktuple() etc are recursive, so nested borrowed references would still be released before the outer do_ignore() function releases the outer tuple.