Re: Conditional property validation

From: David Abrahams (dave_at_[hidden])
Date: 2004-05-27 07:32:10

Vladimir Prus <ghost_at_[hidden]> writes:

> Hi Jürgen,
>
> Jürgen Hunold wrote:
>> lib boost_test_exec_monitor : $(TEST_EXEC_MON_SOURCES).cpp
>>
>> : <toolset>borland:<cxxflags>-w-8080
>>
>> ;
>>
>> This breaks with the following message:
>>
>> error: "borland" is not a known value of feature <toolset>
>> legal values: "gcc"
>>
>> when no "using borland ;" is specified in user-config.jam or elsewhere.
>>
>> I think this is a know issue, isn't it ?
>
> No, this is not a known issue and it fact I've introduced the regression (?) 
> yesterday. Until now, the properties inside condition (the <toolset>borland 
> part above) was not checked for correctness at all. While fixing another bug, 
> I've accidentally caused it to be checked.
>
> Now the question:
>
> Do we need validation for condition?
>
> For example, if we have:
>
> <toolset>msvc:<cflags>...
>
> it's possible to ask user to have "using msvc ; " somewhere. But that might 
> cause warnings that msvc is not found. Further, for
>
> <toolset>msvc-7.1:<cflags>...
>
> user should declare *specific* version of toolset.. It's hard to do that 
> inside boost's project-root.jam, because it's not clear what path to use, for 
> example, and how the user can configure the same version in user-config.jam
>
> So, we've two choices:
>
> 1. Don't validate condition at all
> 2. Provide the means for user to say: 'msvc-7.1' is OK as the value. This, 
> however, might be very tricky, since "msvc-7.1' is really a shortcut for

3. Automatically do "using msvc" the first time such a condition is
encountered. However, set things up so the toolset won't complain
about not finding the compiler.

-- 
Dave Abrahams
Boost Consulting
http://www.boost-consulting.com