On 06/13/2010 12:07 AM, Ralf Wildenhues wrote:
Hi Peter,
* Peter O'Gorman wrote on Sun, Jun 13, 2010 at 06:07:44AM CEST:
On 06/11/2010 11:56 PM, Gary V. Vaughan wrote:
1: There might be a better way. I'm thinking...
I'm not sure what testing that modules cannot be unloaded gets you
when you don't have shared libraries.
That may be true, but the failure was during lt_dlopen. I think it
should be possible to preload modules with the resident bit, no? I
mean, they are the prototype of resident modules, their closing is what
won't do the right thing for non-residents (and maybe a future
improvement of the preopen loader could even emulate that).
This failure looks like a genuine ltdl bug to me; why not leave it open
for now?
Hi Ralf,
IIRC this one could also be fixed by using LTDL_SET_PRELOADED_SYMBOLS, I
don't remember for sure, but I recall being convinced that these were
all testsuite issues.
3: eww!
It's either skip the test entirely or some crap like this, you can't
load a static archive "RTLD_LOCAL", then load a different static
archive with the same symbols later "RTLD_GLOBAL", and expect things
to work. I debated what to do, and decided to do the crap thing,
since it's possible some parts of the test are useful.
I don't have a better idea here either; only that something like your
paragraph above should be a comment right before the hack, so we don't
ask ourselves the same question again next time.
Ok, I'll try to come up with a comment that doesn't use the word 'crap' :-)
Peter
