Re: [Boost-testing] IBM tests

From: Rene Rivera (grafikrobot_at_[hidden])
Date: 2007-06-29 13:01:52

Christopher Cambly wrote:
>> I see some of the first results for the IBM regression testing at 
>> <http://beta.boost.org:8081>. I have one request, it would be good to 
>> configure the toolsets so that they include the version of the compiler. 
> 
>> So instead of "vacpp" perhaps "vacpp-9.0".
> 
> We would prefer something like ibmxl-9.0 since the compiler has not been
> branded "Visual Age" for quite some time. However, that appears to require 
> more changes than seems to be implied above. 

Yea, that would require changing the toolset file from vacpp.jam to
ibmxl.jam. Not that I would be against such change, but is there a good
name that covers both the old and new names? Perhaps the internal name 
of the compiler, IBMCPP? So maybe naming it ibmcpp.jam?

> We have much to learn about the regression reporting mechanism, and the 
> Boost development process in general.  How do we make the change you 
> suggest?

You'll need to create a user-config.jam, or test-config.jam, and 
specifically configure the toolsets you want further specifying the 
version and any other options (note, the version can't have "-" in it). 
For example take a look at my user-config.jam which defines custom 
versions for the gcc compilers 
<http://engineering.meta-comm.com/boost-regression/CVS-HEAD/RSI%20Droid.html>.

> I was surprised by several things during our initial run for RC_1_34_0
> and CVS_HEAD.  First, the regression.py that was downloaded from meta-comm
> failed to upload results for RC_1_34_0 with an error that collect_logs
> takes 9 args, but 10 were specified. 

That would be because of a mismatch between the version in HEAD vs. the 
RC_1_34_0 one. Since the HEAD version includes my changes for Dart it 
has some extra args in places.

> I collected the logs manually.

Yes, that also works since it's just a difference between regression.py 
(the frontend script) and collect_and_upload_logs.py (which parses and 
uploads the results).

> It 
> appears that regression.py is connected to the CVS release on which the
> regression testing was done.  Is this correct?

Yes. For 1.34.x you should use the regression.py in the RC_1_34_0 
branch. You can download it directly from CVS with something like:

wget 
'http://boost.cvs.sourceforge.net/*checkout*/boost/boost/tools/regression/xsl_reports/runner/regression.py?pathrev=RC_1_34_0' 
-O regression.py

> I also was surprised that the RC_1_34_0 regression reports appeared on
> <http://tinyurl.com/lawq3>

Look at your platform description you have a 24 CPU machine... must be 
nice ;-) Anyway, if you use the latest bjam from CVS HEAD you can then 
use the "-j" option to make use of some of those CPUs. The tests Noel 
runs on CVS HEAD do that, for example 
<http://engineering.meta-comm.com/boost-regression/CVS-HEAD/Sandia-darwin.html>.

> but the CVS_HEAD results appeared on DART.  I 
> am
> reading though the archives to find out more information about DART.
> Do you have documentation we can read?

<http://www.na-mic.org/Wiki/index.php/Dart2Summary>

> In addition, we noticed that some of the patches for BBV2 on AIX in Trac
> #583 that were applied to RC_1_34_0 were not merged to CVS_HEAD, 
> specifically
> related to python.

Yea, the difference between 1.34 and HEAD Boost.Build are aggravating 
:-( I tried merging some changes back to HEAD but the way some of the 
changes got moved around makes it very difficult. And I don't know 
enough about what the changes are to really know how to merge. I asked 
Volodya to look into it, but I guess he hasn't had time.

-- 
-- Grafik - Don't Assume Anything
-- Redshift Software, Inc. - http://redshift-software.com
-- rrivera/acm.org - grafik/redshift-software.com
-- 102708583/icq - grafikrobot/aim - grafikrobot/yahoo