Re: [libvirt] [PATCH] util: portably check for unchanged uid
On 03/15/2013 10:47 AM, Laine Stump wrote:
On 03/14/2013 06:35 PM, Eric Blake wrote:
> We've already scrubbed for comparisons of 'uid_t == -1' (which fail
> on platforms where uid_t is a u16), but another one snuck in.
ACK (and sorry for the botched "bugfix" :-/)
> +# Don't compare *id_t against raw -1.
> +sc_prohibit_risky_id_promotion:
> + @prohibit='\b(user|group|[ug]id) *[=!]= *-' \
> + halt='cast -1 to ([ug]id_t) before comparing against id' \
> + $(_sc_search_regexp)
> +
As we discussed on IRC, I'm slightly concerned about false positives
when user, group, or [gu]id is used as something other than [gu]id_t,
but the most common case of this would be for a char*, and I doubt we
would ever be comparing a char* with !=,
Well, there's always 'group == NULL' or 'group != NULL'. But neither
of
those patterns match the '-' of '-1' on the other side of the
comparison, so they won't flag as a syntax error.
so I think it's okay (certainly
better than the alternative of pushing a release only to find that we
don't compile on some platform).
Actually, it's worse than not compiling - it's a silent difference in
behavior. Omitting the cast has a valid compilation regardless of the
size of uid_t, it's just that the rules for how C99 requires it to
behave are not intuitive to a naive programmer who doesn't realize that
the size of uid_t plays a role in what behavior results, and that the
behavior is different for 32-bit vs. 16-bit types. If we could actually
make it a compiler error (such as with a -Wfoo -Werror combination),
that would be nicer than a syntax check, but as far as I know, gcc
doesn't have anything to help us out on that front.
One thing I did not check for is yoda-isms like '-1 == uid'; but as we
already actively discourage that coding style, I don't think we have to
worry about missing any of those violations.
Thanks for the review, and pushed.
--
Eric Blake eblake redhat com +1-919-301-3266
Libvirt virtualization library http://libvirt.org
Attachments:
- signature.asc (application/pgp-signature — 621 bytes)