[Baypiggies] A talk on testing frameworks

[Minesh B. Amin]

Coming at this problem (of testing distributed/parallel software)
from the perspective of "parallel processing", we finally realized
that a host of issues (including the ones Lincoln mentioned)
can be much better described, prototyped, implemented, tested, and 
studied by going back to basics in terms of how package management 
is done.

Our solution had to have couple of attributes:
  + pkg management must be embedded within the Python interpreter 
  + pkg dependencies must be very easy to describe, compute (if needed),
    enforce/freeze, and migrate from one snapshot to another 
  + must be very easy to describe, and run/collect stats on multiple
    runtime environments in parallel (with each having a different
    set of pkg dependencies)
  + must be able to install at any time, potentially conflicting,
    versions of the same pkg under some "install directory" ... 
    without the addition causing any problem

If interested, I can give a presentation on the what/why/how of
our solution.

Cheers!
Minesh
 
On Sun, 2011-01-16 at 12:07 -0800, Lincoln Peters wrote:
> On Sat, Jan 15, 2011 at 1:49 PM, Glen Jarvis <glen at glenjarvis.com> wrote:
> > What interest is there and what are people most curious about?
> 
> A few things that I'd be interested in:
> 
> * Controlling multiple hosts for a single test
> * Measure and report how long it takes to run a particular test, and
> compare to previous builds
> * Save artifacts, so that a stakeholder can visually inspect output
> for correctness
> 
> At my job, we use the built-in "unittest" framework with a few of our
> own derived classes for most of our testing, but I haven't yet found
> good solutions to those problems.
> 
>