Hi Greg,
Not sure if this is off-base, but is it correct to assume that __wine_pop_frame is based on setjmp/longjmp? If not, ignore the rest of this note.
If it is, then one issue to think about is that all the user variables in the handler blocks which are also used in the try block must be declared volatile. For example,
int x; try { ... x = ... } handler { ... y = x; }
won't work right unless x is volatile (because the compile's flow analysis and register usage will get confused). Since the original code is unlikely to be so marked, you have a problem. In general, AFAICS, any solution implemented at the language level has this problem...without compiler support you are stuffed.
I am in the middle of solving a similar problem at work (i.e. forward porting and old C codebase using both exceptions, and a homebrew ORB which also has exceptions handled in a unified way). The ONLY solution I could come up with was indeed to use C++'s native exceptions, but wrapped in macros in such a way that the legacy code works without change. In fact, it works better than the legacy code because the legacy coders sometimes forgot to mark variables as volatile :-).
If this is any use to you, contact off-list (I am not subscribed), and I can assist with the conceptual version of the macros - I've long forgotten the MSVC exception details.
Thanks, Shaheed
-
Shaheed R. Haque