incubator-stdcxx-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Martin Sebor <mse...@gmail.com>
Subject Re: ACC 3.76 test builds
Date Wed, 18 Apr 2007 22:19:04 GMT
Andrew, if you haven't done so yet, could you please forward
this useful info (or a send a link to the archived post) to
the HP aCC folks?

Thanks
Martin

Andrew Black wrote:
> Greetings all.
> 
> The nightly testing system is somewhat backed up with pending builds for 
> other (RogueWave) products at the moments.  The platforms with the 
> greatest backlog are HPUX 11i and 11iv2, with Solaris and Windows having 
> somewhat shorter backlogs.
> 
> One compiler we have been attempting to test on HPUX is the aCC 3.76 
> (beta) compiler.  This testing has been hampered by a couple problems. 
> One problem was an abundance of warning messages, leading to the console 
> log being larger than size allowed by the nightly testing system.  This 
> has been partly solved by suppressing some of the most obnoxious 
> messages.  The second problem is a recently diagnosed flaw in the 
> nightly testing system which cleans out failed builds much sooner than 
> it should have.  An update to the nightly testing system is expected to 
> be deployed in the near future, and hopefully will resolve this issue.
> 
> Moving beyond these couple issues related to the nightly testing system, 
> I have observed a couple areas of failures in a manual (11d) build.  One 
> failure is the localedef utility entering what appears to be an infinite 
> loop in the constructor of a Charmap object.  This is reminiscent of a 
> similar problem observed on windows, though I haven't stepped through at 
> the level necessary to confirm this observation.  This condition also 
> leads to the exec utility killing the locale tests after the build 
> window expires.  However, the cumulative duration of these build windows 
> is greater than the build window allowed by the nightly testing system 
> for any particular run, leading to the build being killed by the nightly 
> testing system prior to completion, which is logged as a failure.
> 
> Also observed in manual testing is the troubling situation of many 
> examples terminating with a SIGBUS condition.  Examples which didn't 
> terminate in this condition generally produced different output from the 
> expected or segfaults.
> 
> With a little luck, the nightly testing system will produce builds with 
> this compiler in the next day or two, so that they can be included in 
> the nightly testing page.
> 
> --Andrew Black
> 


Mime
View raw message