Re: RFC: proposal for indirect deplibs

[Bob Friesenhahn]

On Thu, 25 Nov 2004, Joe Orton wrote:

> On Wed, Nov 24, 2004 at 10:14:35PM -0800, Noah Misch wrote:
> > For safety, libtool would assume that unadorned dependent libraries contribute
> > to the API.  An option, say `-private-libs', would mark the start of normal
> > dependencies.  For example, the link command for libbar might look like this:
> >
> > libtool --mode=link cc bar.o... -o libbar.la -lfoo -private-libs -lm
> >
> > Consider baz, an application that links with libbar.  In Ralf's terms, libfoo is
> > then a direct base library of baz, and libm is an indirect base library.
>
> I like this a lot, I think it's pretty important that the determination
> of which libraries are inherited as "direct" and which as "indirect" is
> left to the library author and not magically guessed by libtool.

Maybe so, but the specification of "indirect" libraries needs to be on 
a per-library basis.  It is wrong to assume that libraries much 
further to the right on the link line are not dependended on directly 
by a library listed very early in the library list.  Necessary/ideal 
library ordering is usually determined by mutual dependencies.  The 
world order may be modified by a single #define.

Unfortunately, a library which is "indirect" to one library may be 
"direct" to another.  When libraries are used in conjunction with each 
other, it may be necessary to treat a library differently than if it 
was not used in conjunction with another library.

I continue to fear that Pandora's box is being opened, and that 
important decisions will be made by people who are actually clueless 
about how the software works, and how it is put together.

Bob
======================================
Bob Friesenhahn
address@hidden
http://www.simplesystems.org/users/bfriesen