On Tue, 1 Feb 2022 at 12:03, Florian Weimer <fweimer(a)redhat.com> wrote:
* Jonathan Wakely:
> Vitaly, it looks like you didn't respond to this. I'm also curious why
> this change would lead to crashes. Are we missing something?
I've seen cases where access to uninitialized data was fine as long as
the memory location was never zero, something that was always true for
how GCC compiled the program at the time.
Ah, so uninitialized pointers that were non-zero, and so reading from
some arbitrary mapped page. If the pointer gets initialized to zero
reading from it would be a segfault, because the zero page isn't
mapped.
That seems like an improvement, and worth finding and fixing the code.
"Maintainers should not have to fix bugs in their packages" seems like
a totally bogus argument to me.
But I most say that I find the other direction more likely (as in, the
program is fine because it works correctly on Fedora).