[omniORB] Incorrect behaviour handling exceptions in VC++ 7.1

[Michael Kilburn]

Hi,

I have discovered a fix in recent version of omniORB that is described as:

Incorrect behaviour handling exceptions on VC++ 7.1 (bug number 2)

Where can I find more detailed info on this, like:
- what exactly is wrong with VC7.1?
- which code change was made to workaround the problem?

[Duncan Grisby]

Post by Michael Kilburn
Incorrect behaviour handling exceptions on VC++ 7.1 (bug number 2)
- what exactly is wrong with VC7.1?
- which code change was made to workaround the problem?

The bug was reported to me directly, so there's no public report of it,
I'm afraid.

The problem is that VC++ 7.1 generates incorrect code for the code that
invokes operations. The old version of the code used continue within an
exception handler. That caused VC++ to generate invalid code so that if
an exception occurred, the runtime state was corrupted. The modified
version of the code uses a flag to represent the same condition, and
therefore does not trigger the VC++ bug. omniORB's code was correct --
the bug is in VC++.

If you want to see the details of the change, it's in CVS in the file
src/lib/omniORB/orbcore/omniObjRef.cc, between revisions 1.4.2.5 and
1.4.2.6.

Cheers,

Duncan.

[Michael Kilburn]

Post by Duncan Grisby
The old version of the code used continue within an
exception handler. That caused VC++ to generate invalid code so that if
an exception occurred, the runtime state was corrupted.

Hmm... I can not reproduce it in small test app.
1. Which state gets corrupted? I.e. how do you know that something went
wrong?
2. Is this VC7.1 bug well-known? (i.e. described in official sources)
3. What exception handling model you use when compiling omniORB with VC7.1?
(/EH option)