On 10/15/20 12:05 PM, Eli Zaretskii wrote:
> I want to continue calling doprnt directly, not replace all its
> calls with a call to another function.

OK, then attached is a patch that does things that way. This patch affects only 
the implementation of doprnt; no doprnt callers are affected. The code should 
have most of the proposed performance benefits of the earlier patches I proposed.