Re: [Boost-build] Method for expressing complex requirements logic

From: Phillip Seaver (phil_at_[hidden])
Date: 2006-01-28 23:27:11

Vladimir Prus wrote:
> Hi!
> Some of the Boost libraries have pretty contrived logic in their regression 
> tests. So example, if the toolset is borland, the serialization library will 
> check for global variable SPIRIT_ROOT, and if that's set, add its value to 
> <include>, and if it's not set, error out.
>
> In V1 this is accomplished by adding a name of a rule to requirements, which 
> rule will be called to compute extra requirements.
#1 (allow rule names in requirements list) sounds the easiest from a
user perspective at first glance.

For #3 (custom replacement for generator.run), it depends on what would
have to go inside the rule.  If I wanted to do a custom executable
build, would I need to copy linking-generator.run, or would we just call
it after making whatever changes to the parameters we needed?

If it's just a case of calling the run rule for the generator that
should have been called, they look about the same, except that you can
modify the sources with #3.  If you don't need to do that, #1 sounds
better, because then you still have the target name to tell you at a
glance what it's going to build ("lib", "exe", etc.) instead of having
to read through the rule's code.

Phillip