Hi, On Sun, 17 May 2009, Michael Eager wrote:
> If the LSDA is only interpreted by the personality routine pointed to > by the unwind table, then all that should be needed is to describe the > the functionality of that routine. Yep, that was what my confusion above was about, if the LSDA format specifically should also be described. I consider describing it in an ABI in a normative part to be actually harmful on the grounds that it's not necessary but would prevent future changes of the format (without ABI changes) ... > Itanium C++ ABI on the CodeSourcery site.) The details on how to generate > the LSDA would likely be an appendix showing an example implementation. ... whereas this of course would be something useful. > The ABI would also need to say something about the name or linkage of > the personality routine, so that if different compilations have > different routines, they won't collide. > (The AMD64 ABI says that there is no psABI-specified name, but that > doesn't prevent collisions.) Indeed. The question is if the ABI should try to prevent such collisions (it would have to list all existing current symbols and be amended whenever someone invents a new one). I'm undecided. Ciao, Michael.
