On 4/2/2011 4:35 PM, Guenter Knauf wrote: > Am 02.04.2011 20:29, schrieb William A. Rowe Jr.: >> On 4/2/2011 10:27 AM, Guenter Knauf wrote: >>> *if* we want to do it 'right' in one pass then I believe we would need to >>> build with >>> APR_DECLARE_STATIC and then get the DDL linked with a .def file (which is >>> close to what is >>> done on Linux with the export*.c hacks); >>> I would volunteer to modify our make_nw_exports.awk to be able to write a >>> Windows .def >>> file, but I would fail to integrate this into a configure build process ... >>> :-( >> >> You know that leads to segfaults, and per-process modification of what are >> otherwise >> constant/shared .text pages, right? > nope - I didnt yet ...
When the compiler has the insight that it is binding against relocatable, loadable sections, it has a chance to change all references to a vtable in mutable heap, and leave the actual binary images alone. Failure to do this for data pointers is usually lethal. The inverse of what you suggest, compiling it all export whether it will be used statically or dynamically works out a bit better, but it's still worth performing two compilations.
