rerunning payday throws off nparticipants

[chadwhitacre]

The paydays table accumulates nparticipants while payday is being run. If the run dies and we rerun it, we end up double-counting people. Surfaced on #204.

[zbynekwinkler]

We do not show nparticipants anywhere now. One option might be to just drop it. Or if we decide to keep it, why wouldn't clearing to 0 in zero_out_pending (at the start of the payday) solve the problem? BTW: the same problem is with ntippers and ntips.

[chadwhitacre]

We show nparticipants on the charts page. Here's what it looks like right now, after Gittip 85 (#1913):

We should reset the nparticipants count for the current payday when we run it.

[chadwhitacre]

Here's the fix:

update paydays set nparticipants=(select count(*) from participants where claimed_time < ts_start);

[zbynekwinkler]

Isn't it easier to just add this query to the end() of payday than trying to find the issue about it, not being able to, creating new, finally finding it, closing the new... 😉

[chadwhitacre]

That's a hacky fix, no? This updates all paydays, which is ugly to do every time, whether or not we need it. Better to fix this properly by ... whatever it would take. :-)

/me out of CPU

[zbynekwinkler]

It is easy enough to update just the last one. And as for

whatever it would take

let me disagree. We should spend as little time as possible on such a minor
almost-bug. Just tracking this in the issue tracker is taking everyone's
"CPU time" that could be used for something more worth while. You know what
they say, choose your battles wisely.

[chadwhitacre]

From #1915:

We should reset the nparticipants count for the current payday when we run it.

[chadwhitacre]

I started working on this, but if I'm going to fix this in code that means writing a test. This happens infrequently enough that I think that manually repairing is still the path of least resistance. We should be looking at ★★★ and DevX ★ tickets when we browse tickets, and this is neither.

[zbynekwinkler]

What I am trying to say is that just tracking this problem here takes more
"CPU time" than fixing it. If the need to make a test for it is the cause
why it is not fixed yet, well... I'd rather have it fixed without a test
and gone from the issue tracker so it never ever gets in the way when
everyone works with the issue tracker. But that is just IMHO.

[Changaco]

This was fixed in the payday rewrite (#2508).