Re: [boost] Regression tests: how long is reasonable?

From: William E. Kempf (williamkempf_at_[hidden])
Date: 2002-08-12 16:10:29

----- Original Message -----
From: "Beman Dawes" <bdawes_at_[hidden]>
> At 04:24 PM 8/12/2002, Gennadiy Rozental wrote:
>  >2. On some (UNIX) systems test case timeout could be set right now. At
> some
>  >point I will try to tuckle windows timeouts.
>
> I've already got contributions of Win32 timeouts from two people. But
> nowadays, we've got Boost Threads, so we might think of a generalized
> approach.  The only trouble with that is the dependency issue; since we
use
> Boost.Test to test Boost.Threads, it may not be a hot idea to make
> Boost.Test dependent on Boost.Threads:-(

As long as the Boost.Test stuff were factored into optional extensions I
could just avoid using these extensions in Boost.Threads testing, thus
breaking the dependency cycle.  So I don't think that's a compelling issue.

However, I'm not sure that what's need by Boost.Test would rightfully belong
in Boost.Threads.  Going from memory, what's needed is a mechanism to abort
a thread immediately if it takes too long to complete.  This is basically an
asynchronous cancel, and I think most people are in agreement that that
concept is to wrought with peril to be useful.  I can see living with the
dangers in tightly controlled unit tests (i.e. there needs to be no shared
resources that we can't safely abandon, or the test run must terminate() on
such failures), but not in a generalized library.

Bill Kempf