Re: [Boost-testing] Build monitor killed my test?

From: David Abrahams (dave_at_[hidden])
Date: 2007-02-04 21:42:39

Bronek Kozicki <brok_at_[hidden]> writes:

> David Abrahams wrote:
>
>> I'm really sorry, but I just don't have the time to do anything like
>> that right now.  I am currently passing -lx120 to bjam and *not* using
>> the "monitored" test option.  I cc'd you because I suspected you might
>> be the person who "wrote that part of the code," due to your expertise
>> in this area.  Only crashed tests produce these dialogs.  I'm doing
>> release mode testing.  libs/iterator/test/filter_iterator_test is one
>> of two that produce the dialog for me.  I hope that gives you some
>> information to go on.
>
> the code to close alerts is "only" few months old, so if you are using 
> older build of boost

You mean bjam?  It was built within the last few weeks; but I'll build
a new one.

> you need new build. Now, the code has serious 
> limitation - it's unable to close windows displayed on desktop that is 
> different from the one bjam is runnig on. One might thing "surely all 
> asserts are displayed on the very same desktop as program that crashed" 
> but this would be mistake. Assertions are (often - depending on 
> compiler) dialog boxes displayed with option MB_SERVICE_NOTIFICATION or 
> MB_DEFAULT_DESKTOP_ONLY, which (through API magic) would display dialog 
> on interactive desktop. If bjam is not running on interactive desktop 
> (eg. because of screen saver or fast user switching or terminal 
> sessions), 

I'm running my tests from a .bat script that gets run as a "Scheduled
Task."

> it won't be able to find and close assertion alert window - 
> because these two are different desktops. This is because of Windows 
> limitation - no one, not even administrator can reach windows on desktop 
> other than own. Well, I think there is way around - SYSTEM user probably 
> would be able to, but that requires rewriting whole functionality as a 
> system service. Or running bjam as SYSTEM and putting a little extra 
> code in there to iterate over desktops.

Well, I'm afraid we really need that "way around" if we're going to
have effective automated testing on Windows.  Otherwise a crashed test
stalls the whole process; a new test run won't start until all those
dialogs are closed.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com