Recent msi/package tests failures

[James Hawkins]

On Thu, Aug 28, 2008 at 7:47 AM, Paul Millar <paul at astro.gla.ac.uk> wrote:
> On Wednesday 27 August 2008 21:15:48 James Hawkins wrote:
>> That's not why the tests are failing.  The install tests are timing
>> out, and if the winetest executable kills the child process that it
>> believes is 'hung', then you're killing the installer process midway
>> through an install and thus leaving the system in a broken state.
>
> Would it make sense to have a generic mechanism for cleaning up after a
> time-out?  For example, if the child executable accepted an argument
> (e.g., /cleanup) that would tell the test to undo any partial state it might
> have established.  For example, it could remove temporary files and some
> known-test registry entries.
>

That's near impossible to do for the install tests.

> Should a child process timed-out, the parent process could call the child
> telling it to clear up any mess before continuing with the next test.
>
> I guess timeouts are always possible, it would be nice if the subsequent
> test-runs were unaffected by this.
>

Or we just increase the timeout for legitimately long tests.  There's
no way to trivially clean up after the install tests.  The reason we
have timeouts is to kill 'stuck' children, but that's not the case
here.  The reason some people's machines are in a bad state now is
because an install was killed halfway through.

-- 
James Hawkins