[PULL] Kill libxorg

Daniel Stone daniel at fooishbar.org
Tue Jun 28 07:28:33 PDT 2011


Hi,

On Thu, Jun 23, 2011 at 05:52:47AM -0700, Dan Nicholson wrote:
> Having now played around with trying to kill libxservertest, I
> realized that this patchset isn't quite as robust as I thought. It
> turns out that getting all the Xorg convenience libraries to link
> without putting them into a massive convenience library isn't strictly
> possible. Code in libcommon uses symbols from libxorgos and vice versa
> for example, so there's no way to order the libraries correctly at
> link time to make the symbols resolve.
> 
> The only reason this patch works is because of sdksyms.c. Since it
> references every symbol, if it's linked as an object into Xorg, it
> will make all subsequent symbols in the link resolve correctly. The
> point being that if sdksyms.c ever went away, Xorg would stop linking.

Right.  For all its flaws (of which there are so very many), it does
keep things hanging together nicely.

> I think there's another way to keep the speedup that would be more
> robust: keep libxorg.la but make it build all of hw/xfree86
> non-recursively to avoid convenience libraries of convenience
> libraries. That would be really intrusive, though.

Heh.  Feel free. :)

> Anyway, I think what I've done right now is fine for the foreseeable
> future. I just wanted to share my thoughts before this goes in.

Seems fair; killing sdksyms.c would be a large enough endeavour that I
think the benefits of this patchset still strongly, strongly outweigh
the downsides.  I'd be very happy to see it pulled (nudge, nudge).

Cheers,
Daniel


More information about the xorg-devel mailing list