Re: [Boost-testing] regression.py and --incremental

From: David Abrahams (dave_at_[hidden])
Date: 2007-02-04 21:32:01

Misha Bergal <mbergal_at_[hidden]> writes:

> Martin Wille wrote:
>> David Abrahams schrieb:
>>> David Abrahams <dave_at_[hidden]> writes:
>>>
>>>> When I pass --incremental I always delete the old bjam.log first because
>>>> otherwise regression.py will just append to the existing one.  Am I
>>>> doing the right thing?
>  > AFAICS, deleting bjam.log is not the reason for the holes. The reason
>  > usually is some problem with parsing bjam.log in order to produce the
>  > XML files.
>
> Martin is completely right. At the end, whether you delete it or not 
> does not matter too much. 

Well, the holes seem to be gone now, FWIW.

> Here the simplified explanation of what is 
> going on:
>
> process_jam_log analyzes bjam output and extracts compile, link and run 
> actions results and output. 

Yes.  For tests that passed the last time around, it will not find
anything to extract if I clear bjam.log

> The result and output for each found action 
> is stored in .xml file in the target directory of test or library (where 
> its .obj, .exe and other files are). If .xml file already exists there, 
> it gets loaded, processed action overrides the corresponding section and 
> saved in the target directory again.

Maybe an aborted test run would cause the holes, if I stopped it
before all the xml files were done being created?

> After process_jam_log is done, regression.py searches whole bin.v2 
> directory for those .xml files (test results), packages them into one 
> big file, zips it and uploads to meta-comm.com
>
> In case of incremental run only the small number of those .xml 
> files/test results are updated, but all of them still get collected by 
> regression.py and sent to us.
>
> Deleting of bjam.log just saves you some process_jam_log processing time 
> and should not change any results.

Hm.  Thanks for the explanation.

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