On Thu, Nov 29, 2007 at 02:57:50PM +0100, Alexandre Julliard wrote:
Dan Hipschman dsh@linux.ucla.edu writes:
In this revision, I leave setting permissions on page faults to the unhandled exception filter and simply ignore page faults on execution access (read access exceptions should be caught since they translate into bad stub data, but these don't pose a problem anyway). I also fixed a bug regarding how the mask field of __widl_except_frame was set.
There's no reason to ignore execute access errors, those are bugs too. You are still not filtering the exception properly though, the condition checking has to be done directly in the handler, you can't longjmp back to the code and then re-raise the exception. That's doubly true for the finally handling. This means you have to generate different handlers for the different exception conditions.
That's basically what I did in the original patch using __TRY/__EXCEPT/__FINALLY. You told me to do this in a more straightforward way:
http://www.winehq.org/pipermail/wine-devel/2007-November/060515.html
which I took to mean trying to implement the finally block as an actual block in the local scope, instead of a function (which is a bit of a pain). I don't see any way to execute the finally block in the local scope without rethrowing exceptions, or using GCC's nested functions. The except condition is easier to put into a handler since the only conditions WIDL generates are pretty simple.
Should I go back to the original patch? How did you want it different?