Re: [Boost-build] Call of interest

Subject: Re: [Boost-build] Call of interest
From: Vladimir Prus (vladimir_at_[hidden])
Date: 2009-05-28 12:01:26

Konstantin Litvinenko wrote:

> Vladimir Prus пишет:
>>>     I was not clear when stating Yes :). The case you described will not
>>> working at least on Windows. Correct process will be using different
>>> naming schema when packaging installation. But personally I don't really
>>> known why should I use different naming schema... May be when build some
>>> install that run from DVD to overcome some DVD filesystem restrictions...
>> 
>> Because if your library should be used by outside world, that outside world
>> should use some name to link to the library.
> 
> Yes, that's true.
> 
> And it's seems not likely
>> the outside world will be missing to compute MD5 hashes of something.
>> One possible solution is pkg-config, on linux, which allows to get an
>> actual name of library given "simple" package name, but that's linux-only.
> 
> Outside word don't need to compute anything. When distribute in binary
> form developer must provide description what lib when to link. To
> integrate to some BBv2 setup one could write:
> 
> lib foo : : <file>../lib/foo-1.0-blah-blah.lib <threading>multi;
> lib foo : : <file>../lib/foo-1.0-blah1-blah1.lib <threading>single;
> 
> and so on. Or may be I am missing something?

Suppose that 'outside' world is Visual C++ developer. If you tell to link
to 

        foo-433423426131231231.lib

he can do that. But if next revision of the library changes name to 

        foo-90909090909090.lib

he might be very, and rightly, upset. I think changes in library names
should be minimized. And then, you have to strip MD5 signature during
install step.

- Volodya