From: Roland Schwarz (roland.schwarz_at_[hidden])
Date: 2007-08-13 07:03:14
Vladimir Prus wrote:
> I thought about your patch and existing code.
> First note, already mentioned on IRC, is that I really don't like
> <name>+<real-name> -- this is fairly confusing.
Me too. (When I remember correctly I never suggested to use this.)
> I do think we should
> be using <tag> -- for ordinary targets the name of target
> is some 'base name' while <tag> is a hook to transform the name
> according to properties. I think the same interface makes perfect sense
> for a prebuilt library too.
I also think it this is much cleaner. What might look a little confusing
now is that presence of the <name> feature alone turns normal in
searched lib. (But this is not new.) <tag> alone (of course) is not
sufficient. (<search> also will do of course.) Hmm, but <name> now looks
a little bit redundant doesn't it?
> Second, while reviewing your patch I noticed that searched-lib-target
> is very crazy -- it has both 'name' and 'real-name' properties, where 'name'
> is not used in any particular way. While fixing that, I've managed to make
> <tag> work for searched lib target. The patch is attached.
Looks fine to me. (Altough I had no time yet to try it out.)
> I also attach test Jamroot that *almost* works -- we only need to strip 'lib'
> from the library names, as 'libboost_program_options-NNNN' is right name
> for created library, but we need 'boost-program_options-NNNN' on
> the linker command line.
> What do you think?
As you might see from my suggested patches:
1) I modified the types, so SEARCHED_LIB will not generate the lib
prefix (which I think is wrong in the general case).
2) I modified the tag rule to take care for the lib prefix for windows only.
Roland aka speedsnail
Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk