Re: enable_timeout_every() and fin_time

Hi,

On 2023-01-03 13:33:34 -0500, Robert Haas wrote:
> On Sun, Jan 1, 2023 at 7:36 PM Andres Freund <andres@anarazel.de> wrote:
> > What is the use case for an absolute start time plus a relative
> > interval?
> 
> The code snippet that you indicate has the important side effect of
> changing the global variable startup_progress_phase_start_time, which
> is used by has_startup_progress_timeout_expired. Without the fin_time
> argument, the timeout machinery would have to call
> GetCurrentTimestamp() separately, and the caller wouldn't know what
> answer it got. The result would be that the progress reports would
> indicate an elapsed time relative to one timestamp, but the time at
> which those progress reports were printed would be relative to a
> slightly different timestamp.

> Maybe nobody would notice such a minor discrepancy, but I wanted to avoid it.

Doesn't that discrepancy already exist as the code stands, because
startup_progress_phase_start_time is also set in
has_startup_progress_timeout_expired()? I realize that was an example, but the
issue seems broader: After the first "firing", the next timeout will be
computed relative to an absolute time gathered in timestamp.c.

Greetings,

Andres Freund