Re: [Boost-build] Alias defect report + test - revoking this report/patch

From: Jurko Gospodnetiæ (jurko.gospodnetic_at_[hidden])
Date: 2008-01-04 13:05:14

   Hi.

>   I believe that when an alias alternative matching the current 
> feature-set can not be found that Boost Build should not display a 
> warning to that affect unless someone actually wants to use (build/use 
> as source/whatever...) that alias with that feature set.

   After looking at this problem a bit more, I am no longer certain that 
there is an easy fix for this.

   Making all aliases explicit by default does not seem to be ok since 
we might actually want an alias built implicitly (e.g. having some 
otherwise explicitly built targets be build implicitly but with some 
additional properties).

   I considered making them explicit only if they have no sources 
defined. That solution starts getting complicated when you consider that 
an alias might have dependencies with sources and/or may have multiple 
alternatives, some with sources and some without.

   Another use case for such a rule would be prebuilt libraries which 
could be made explicit, but that case also becomes difficult to solve in 
general (again, think multiple alternatives, some prebuilt and some not).

   In the end I guess having to mark a target as explicit in these cases 
does not seem like such a bad solution. :-)))

   I have some ideas about allowing a specific alternative to be 
implicit or explicit, effectively allowing specific alternatives not to 
be included in the search when a target is explicitly requested but I 
have not given this enough thought to know how it would work out. For 
now I added it to my 'to do' list. :-)

>   I'm attaching a bit refactored alias rule tests (split into two files) 
> and an added alias_no_warning.py test for this catching this warning.

   In light of the reasoning above, neither this patch nor the one for 
test_all.py (also posted in this thread) should be applied and I'll post 
a different patch containing all the extra regular 'code cleanup' they 
included.

   Best regards,
     Jurko Gospodnetiæ