On 8/17/06, Bojan Smojver <[EMAIL PROTECTED]> wrote:
Quoting Ruediger Pluem <[EMAIL PROTECTED]>:
> So the pointers to the literals where invalid. Ouch.
Right. I see that my assumption about relying on things after
apr_dso_unload()/apr_dso_load() cycle was incorrect.
I'm not sure how to address that at all. I mean if on Linux pointers
change after dlclose()/dlopen() cycle (with ref. counts at zero),
relying on pointers to be the same is sure to be a bug, right?
I have a feeling that this part of the problem really has to do with
abuse caused by programmers using the API and not APR...
Perhaps, but if we make slight changes to APR's APIs it becomes quite
possible to avoid these sort of problems. It doesn't seem like you
should have to bend over backwards and jump through hoops at the same
time in order to load DSOs inside a library that exposes interfaces
that use pools.
-garrett
