Re: Improving the isolationtester: fewer failures, less delay

Andres Freund <andres(at)anarazel(dot)de> writes:
> Only halfway related: I wonder if we should remove the automatic
> permutation stuff - it's practically never useful. Probably not worth
> changing...

Where it is useful, it saves a lot of error-prone typing ...

> Minor suggestion: I think the folliwing would be a bit easier to read if
> there first were a list of markers, and then separately the longer
> descriptions. Right now it's a bit hard to see which paragraph
> introduces a new type of marker, and which just adds further commentary.

OK, will do. Will act on your other cosmetic points too, tomorrow or so.

>> + if (step_has_blocker(pstep))
>> + {
>> + if (!(flags & STEP_RETRY))
>> + printf("step %s: %s <waiting ...>\n",
>> + step->name, step->sql);
>> + return true;
>> + }

> Might be a bug in my mental state machine: Will this work correctly for
> PSB_ONCE, where we'll already returned before?

This bit ignores PSB_ONCE. Once we've dropped out of try_complete_step
the first time, PSB_ONCE is done affecting things. (I'm not in love
with that symbol name, if you have a better idea.)

regards, tom lane