Re: [Boost-testing] Very strange regression results for borland

From: Bronek Kozicki (brok_at_[hidden])
Date: 2007-02-06 02:29:25

AlisdairM wrote:
>> It seems as if borland test builds are failing, depending on how
>> much other toolsets are specified on the same bjam cmd-line.
>>
>> One of my last runs (speedsnail-dbg) had 7 sets, and borland
>> failed (e.g. serialization).
>>
>> I am currently breaking the regressions runs into smaller sets,
>> grouped by vendor. So both borland results can be seen at the
>> moment.
>>
>> Has anyone a plausible explanation for this behaviour?
> 
> No, but I am running my tests per compiler as separate command lines,
> and then processing the combined log at the end.  Would that help?

I noticed this before - it's bjam consuming huge amounts of memory, 
proportional to number of selected toolsets

B.