Re: psql's \watch is broken

On Mon, Dec 16, 2019 at 11:40:07AM +0900, Michael Paquier wrote:
> As the concepts behind cancel_pressed and CancelRequested are
> different, we need to keep cancel_pressed and make psql use it. And
> the callback used for WIN32 also needs to set the flag. I also think
> that we should do a better effort in documenting CancelRequested
> properly in cancel.c. All that should be fixed as of the attached,
> tested on Linux and from a Windows console. From a point of view of
> consistency, this actually brings back the code of psql to the same
> state as it was before a4fd3aa, except that we still have the
> refactored pieces.

Merging both flags can actually prove to be tricky, as we have some
code paths involving --single-step where psql visibly assumes that a
cancellation pressed does not necessarily imply one that succeeds is
there is a cancellation object around (ExecQueryTuples, tuple printing
and \copy). So I have fixed the issue by making the code of psql
consistent with what we had before a4fd3aa. I think that it should be
actually possible to merge CancelRequested and cancel_pressed while
keeping the user-visible changes acceptable, and this requires a very
careful lookup.
--
Michael