Re: [Boost-build] Disabling library build on a single platform

Subject: Re: [Boost-build] Disabling library build on a single platform
From: Jeremiah Willcock (jewillco_at_[hidden])
Date: 2012-04-16 16:41:02

On Mon, 16 Apr 2012, Vicente J. Botet Escriba wrote:

> Le 16/04/12 21:45, Jeremiah Willcock a écrit :
>> On Mon, 16 Apr 2012, Vicente J. Botet Escriba wrote:
>> 
>>> Le 16/04/12 21:13, Jeremiah Willcock a écrit :
>>>> On Mon, 16 Apr 2012, Vicente J. Botet Escriba wrote:
>>>> 
>>>>> Le 16/04/12 19:51, Jeremiah Willcock a écrit :
>>>>>       The bug report at 
>>>>> <URL:https://svn.boost.org/trac/boost/ticket/5606> suggests Boost.Graph 
>>>>> disabling its build on Sun compilers.  How do I do that?
>>>>> 
>>>>> Hi,
>>>>> 
>>>>> is Boost.Xpressive essential to Boost.Graph?If yes, stop reading.
>>>>> 
>>>>> If not, I would expect that the parts depending on Boost.Xpresive can be 
>>>>> excluded conditionally. Couldn't this be better that forbidding Sun 
>>>>> users to use Boost.graph?
>>>> 
>>>> It's not essential, and is used in a header that most parts of 
>>>> Boost.Graph don't include by default.  The Sun compiler breaks on many 
>>>> other parts of Boost.Graph as well, though, so I'm not too concerned 
>>>> about portability to it (most of the Sun compiler regression test errors 
>>>> have nothing to do with Xpressive).  The IBM compiler seems to break on 
>>>> PropertyTree as well, which is used for GraphML input.  Those are the two 
>>>> major compilers that fail large parts of BGL; my opinion is just to avoid 
>>>> building the boost_graph library rather than trying to only compile those 
>>>> parts of it that aren't broken.
>>> 
>> It sounds to me like the best solution is just to close that bug as 
>> wontfix.  Is that reasonable?  Are there other libraries that are 
>> auto-skipped on platforms where they don't build?
>> 
> See all the n/a here 
> http://www.boost.org/development/tests/trunk/developer/summary.html. You can 
> do the same thing updating the status/explicit-failures-markup.xml as in
>
> <library name="thread">
> <mark-unusable>
> <toolset name="*como-4_3_3*"/>
> <note author="B. Dawes" refid="10"/>
> </mark-unusable>

That just marks the regression test output, right?  That will still cause 
builds (not in regression tests) to fail on those platforms since the 
library itself does not compile.  It sounds like that latter problem is 
not easy to work around.  I'd prefer not to mark those platforms as 
completely skipped since BGL works partially on them, but some features 
are missing (a lot on Sun, a few on IBM).

-- Jeremiah Willcock