[Bug analyzer/98797] Simpler version of the XFAIL in casts-1.c with proposed solution

[brian.sobulefsky at protonmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98797

--- Comment #4 from Brian Sobulefsky <brian.sobulefsky at protonmail dot com> ---
(In reply to David Malcolm from comment #3)
> The branch in comment #0 now gives a 404, but in any case I had to rewrite
> the store code in gcc 12 to support detection of uses of uninitialized
> values, so any patch is likely bit-rotted.  The new store implementation
> allows for a simpler solution that avoids special-casing, which I've
> committed as the patch above, fixing the xfails.

Hi David. As you requested last year I provided an actual patch file generated
by git. If you remember, I did not have a job then and was just sort of looking
around for projects to get involved in. I was new to the distributed
development thing, as I was mostly just a hobby hacker, and so initially was
not sure how to get you the fix. I submitted the patch in an email chain with
subject "Patch for PR analyzer/98797" around February 9, 20201.

A while after, I deleted the repository, as it was both not the preferred way
of submitting gcc patches anyway and also I had not done any gcc work in a
while, so it was just a severely out of date mirror.

I have not looked at your codebase in a while, so if things have changed that
is fine. The solution I tracked down last year was based on the way you had
everything structured at the time. It was a fun project to try to trace someone
else's project like that anyway.

I know we had also had some discussions regarding constraints on another
thread. I had come up with a preliminary way for the constraints to resolve
addition and some other operations, but that problem became a bit more complex
as we needed to decide how to handle the possibility of overflow and I never
saw a final answer on that question. I had gone down that road due to a bug
found in a run from the people at openssl where the constraint manager was not
able to follow a loop correctly. This had involved your "widening_svalue", and
I was not clear on how that worked, but for starters I got it to understand
basic operations like "svalue+3" or whatever, and then that led us to ask how
to handle the possibility of overflow.

If you would like some targeted assistance with development and bug fixing, let
me know.