Re: [Boost-build] bjam crashing on trunk regression test runner VeecoFTC

Subject: Re: [Boost-build] bjam crashing on trunk regression test runner VeecoFTC
From: Jurko Gospodnetić (jurko.gospodnetic_at_[hidden])
Date: 2012-07-20 06:15:09

   Hi.

> I just noticed today that bjam is crashing with a 0xC0000005 access
> violation on our VeecoFTC test runner that runs the trunk regression
> tests.  I'm not sure when this behavior started occuring, but it appears
> that the last date when the regression tests successfully posted for us
> was July 9th.  I believe the regression test scripts rebuild bjam from
> the trunk each cycle, so I am curious if something has become broken on
> the bjam trunk in the last week or two?

   There have been plenty of commits on the trunk in the mean time. And 
there was also one crash bug that was resolved at about that time (that 
one affected only non-Windows bjam builds, caused by a bug in 
unix-specific code).

   Can you reproduce the crash? And find out more precise steps that 
cause it? E.g. is it crashing on bjam's build, on internal Boost 
Jam/Build tests, or on some concrete Boost library test being run with 
it? Is it random? The bug mentioned earlier was random and was caused by 
multiple parallel build actions - one creating a temporary file and 
deleting it while a the main thread detected the temporary file and then 
was missing some error handling when that file could not be found on the 
system a moment later.

   What exact revision are you reproducing the bug with? Could you try 
to manually take a specific Boost Build revision for this?

   I see your test site uses Windows XP/SP3 and msvc-9.0/SP1. I'm 
running Windows 7 & using msvc-9.0/SP1 for most of the Boost Build/Jam 
development here myself and have seen no unexplained crashes on my end 
recently.

   Could you try running the tests once on the same machine without 
cross-compiling for Windows Mobile 5.0 and without using stdport? I 
doubt any of those have anything to do with the crashes, but it would be 
nice to make sure.

   If you do not have the time for this, I'd be glad to remote connect 
to the system and do some of the research myself, if possible.

   Hope this helps.

   Best regards,
     Jurko Gospodnetić