Paul Eggert <
address@hidden> schrieb am Mi., 17. Mai 2017 um 22:38 Uhr:
On 05/15/2017 02:04 AM, Philipp Stephani wrote:
> - Maybe add verify(INT_MAX >= TYPE_MAXIMUM(in_port_t)) and
> verify(TYPE_MINIMUM(in_port_t) == 0) to make sure that we can use int
> for the port?
The code has been changed so that the port number is no longer stored in
an int so this remark is obsolete now. However, if this issue crops up
later I don't think we need to worry about it for IPv4 and IPv6 ports,
since they're always in the range 0..65535. Emacs already assumes that
'int' is at least 32 bits wide, as per POSIX.
Yes, these static assertions are more for documenting the assumptions that the code makes in a way that actually checks the assumptions.
> - Should there be a guard against the alias violation, e.g. by
> declaring sa and sa1 with __attribute__((may_alias))? Otherwise it's
> UB and the compiler might elide the switch entirely.
Yes, that is easy enough to do and would avoid some unlikely but
hard-to-catch bugs. I installed the attached. Most likely other parts of
Emacs should use may_alias too; do you happen to have any tool in your
toolbox that would find them systematically?
A type-based aliasing sanitizer is currently being added to Clang. Once that is landed I'll run it over the Emacs codebase.