[Bug middle-end/32667] block copy with exact overlap is expanded as memcpy

[ch3root at openwall dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=32667

--- Comment #17 from Alexander Cherepanov <ch3root at openwall dot com> ---
On 2016-04-25 17:34, joseph at codesourcery dot com wrote:
> Or it could simply document an additional requirement on the C library
> used with GCC (that the case of exact overlap works), just as it documents
> the requirement that some functions such as memcpy be provided even in the
> freestanding case.  The compiler and library implementations need to
> cooperate to produce a conforming C implementation.

1. Is glibc community ready to provide such guarantee? The doc[1] 
doesn't currently mention an exception for exact overlap.

[1] 
https://www.gnu.org/software/libc/manual/html_node/Copying-Strings-and-Arrays.html#index-memcpy

2. I agree that it should be documented. Together with the assumption 
that memcpy doesn't clobber errno.

3. The decision affects the whole Free Software community. If memcpy 
with exact overlap inserted be gcc are legitimized then users cannot use 
tools like valgrind to catch such cases in their own code. (And neither 
_FORTIFY_SOURCE nor UBSan seem to catch overlapping memcpy now.) So it 
effectively leads to legitimizing such memcpy in user's code too. Making 
memcpy closer to a simple assignment (C11, 6.5.16.1p3) could be a good 
thing. Maybe even fix C std or POSIX?


BTW llvm bug is here: https://llvm.org/bugs/show_bug.cgi?id=11763 . It 
also contains links to valgrind discussions.