Re: [Boost-testing] regression test does nothing

From: David Abrahams (dave_at_[hidden])
Date: 2006-12-04 13:00:17

Vladimir Prus <ghost_at_[hidden]> writes:

> Aleksey Gurtovoy wrote:
>
>> Rene Rivera writes:
>>> The key portion of the problem is the bjam invocation:
>>> 
>>> >> # Starting tests ("...bjam" --v2 ... gcc ...
>>> ----------------------------------------^^^
>>> 
>>> Because the regression.py is using that syntax to specify the
>>> toolset it doesn't select the autoconfigured toolset. If it had used
>>> the "toolset=gcc" form it would work :-\
>> 
>> Hmm, is there a good reason for this behavior?
>
> I think yes. Otherwise, how can we tell 'gcc' from 'random-foobar'? While
>
>         toolset=gcc
>
> clearly indicates you expect 'gcc' to be a name of a toolset, random
> token on the command like can be anything.

That's a major violation of the principle of least surprise.  So, it
can be anything.  So what?  That's true even when toolsets are
explicitly configured.

Unless I'm missing something, we should (perhaps) issue a short
warning and behave just as we would if the user had written
toolset=gcc.

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