Could you explain a bit or make a sample to reproduce?

@Xqt Try to use replace.py and confirm/reject/modify all diffs without having to wait on the "save" thread. It is impossible, so the replace.py logic is not (at least not fully) async.

T147178#3567967

In T196039#4246659, @matej_suchanek wrote:

T147178#3567967

Well, according to my experience, saving using solve_disambiguation works more smoothly than using replace.py. But maybe I'm not right?

In T196039#4246659, @matej_suchanek wrote:

T147178#3567967

What I would expect from Pwb:

• manual mode (y/n) => async
• always mode (a/-always) => sync

How Pwb behaves mostly:

• manual mode => sync (omg why?)
• always mode => async

How Pwb behaves sometimes (replace.py):

• manual mode => async working not so async as it should
• always mode => sync

In T196039#4245909, @Urbanecm wrote:

@Xqt Try to use replace.py and confirm/reject/modify all diffs without having to wait on the "save" thread. It is impossible, so the replace.py logic is not (at least not fully) async.

Cannot really reproduce it. But I re-implemented cached output from compat in https://gerrit.wikimedia.org/r/#/c/pywikibot/core/+/282818/ and discarded all these callback changes in replace.py (which does not work properly especially for other messages than saving hints) to use the default BaseBot's behaviour. In addition I introduce a saving delay for -simulate option with https://gerrit.wikimedia.org/r/#/c/pywikibot/core/+/607072/ and awaiting put threads beeing done during BaseBot.exit() in https://gerrit.wikimedia.org/r/#/c/pywikibot/core/+/607170/. With these concepts I do not find a sticking async behaviour. A good simulating command could be:

py -3 pwb.py replace -simulate a b -start:! -summary:"Bot: Tests"

Confirm several changes with [y]es and stop with [q]uit then.