Re: [Mesa3d-users] undefined symbol in egl/egl_x11_swrast.so and init problems

[Andreas P. <and...@gm...>]

Hi,


2010/4/5 Chia-I Wu <ol...@gm...>:
> On Wed, Mar 31, 2010 at 05:59:16PM +0200, Andreas Pokorny wrote:
>> According to git bisect the fix of chaning RTLD_LOCAL to RTLD_GLOBAL
>> causes a crash in a different es2/egl implementation.
>> What is bad about explicitly linking to libEGL.so in the egl driver?
>> From my limited point of view it work without breaking anything.
>
> Sorry for the late response.
>
> Could the crash happen to be a version mismatch between libEGL and the
> driver?  There is no version checking done right now..

The crash happens on an imx51 board equipped with a z430 of amd. The
mesa stack is not involved. I am in contact with the vendor about this
issue (more or less it is vacation time here). Even though I have a
stack trace about the issue, I have no idea why it depends on the ld
flags. Its just that at the moment a component that is rather
unrelated to the actual rendering needs a build switch to either
support loading a renderer module that either runs with this or the
other implementation of libEGL.

> As for linking back to libEGL, I was worried about cyclic linking.
> Currently, EGL drivers require back-linking to libEGL
>
> http://sourceware.org/autobook/autobook/autobook_172.html
>
> It is assumed to be a bad design.  I hope to fix this intead of linking
> back to libEGL.

So to make the graph directional again, one would have to strip libEGL
from the support symbols, and provide then in a seperate library. Are
these symbols provided by libEGL and used by the loaded modules
required by all moduels, or is this only needed by swrast?

regards
Andreas