lib-y vs EXPORT_SYMBOL: who wins?

lib-y vs EXPORT_SYMBOL: who wins?

[Rusty Russell]

Various files under lib/ are linked into a .a so they only get linked if
needed.  But many of these functions are also EXPORT_SYMBOL()ed.

This doesn't really make sense: if it's exported it really needs to be
present.  Certain configurations can hit this (lguest uses kasprintf,
and can be a module).

We could do something hacky and try to figure out if any modules need
the symbols, which screws modules built later, but is no worse than a
CONFIG_-based solution.

Or to we just move all the exported functions out of the .a?

Cheers,
Rusty.

Re: lib-y vs EXPORT_SYMBOL: who wins?

[Sam Ravnborg]

On Tue, Sep 25, 2007 at 03:07:44PM +1000, Rusty Russell wrote:
> Various files under lib/ are linked into a .a so they only get linked if
> needed.  But many of these functions are also EXPORT_SYMBOL()ed.
> 
> This doesn't really make sense: if it's exported it really needs to be
> present.  Certain configurations can hit this (lguest uses kasprintf,
> and can be a module).
> 
> We could do something hacky and try to figure out if any modules need
> the symbols, which screws modules built later, but is no worse than a
> CONFIG_-based solution.
> 
> Or to we just move all the exported functions out of the .a?


I am all for killing lib-y altogether. But do not want to sacrify
kernel bloat for this.
hch suggested a bit of Kconfig tricekery and that seems
like an OK plan to me.
But I'm not in a position to do the detective work at the moment -
feed up with oter stuff for now.

	Sam