[postgis-devel] interrupt_* tests

[Sandro Santilli]

On Tue, Jul 07, 2015 at 12:20:11PM -0700, Paul Ramsey wrote:
> Is there some way to make these tests less brittle? I seems like 1 in 5 commits to Debbie end up getting flagged as errors thanks to issues w/ these, and I’ve been able to trip them just by running the tests when my computer is under extra load.

Right, extra load is the key.
Problem is that we don't really know how much time will pass before
PostgreSQL has a chance to run the interruption, but we want to check
that interruption happens _as_soon_as_possible_.

How would you do that ?

I was thinking one way could be to run a known operation and use
that timing as a reference base. Still, if load changes between
the reference sampling and the test run you could still have a
failure.

Have fun !

--strk;

  ()   Free GIS & Flash consultant/developer
  /\   http://strk.keybit.net/services.html