On 30.05.2016 16:56, Peter Krempa wrote:
On Mon, May 30, 2016 at 15:53:08 +0200, Michal Privoznik wrote:
After these regular errors, I still cannot build cleanly:
../../src/conf/domain_conf.c: In function 'virDomainChrPreAlloc': ../../src/conf/domain_conf.c:14109:57: error: potential null pointer dereference [-Werror=null-dereference] return VIR_REALLOC_N(*arrPtr, *cntPtr + 1); ^ ../../src/conf/domain_conf.c: In function 'virDomainChrRemove': ../../src/conf/domain_conf.c:14133:21: error: potential null pointer dereference [-Werror=null-dereference] for (i = 0; i < *cntPtr; i++) { ^~~~~~~ cc1: all warnings being treated as errors
These are obviously false positives. I know some developers here are not very fond of fixing those, so my question is - is it worth the effort or will my patches be NACKed straight away?
Looking at the code I'm not surprised that the compiler thinks that it's a null-deref. The only reason why it isn't is that we don't ever pass the VIR_DOMAIN_CHR_DEVICE_TYPE_LAST as device type and that's very hard to infer.
An option would be to remove the monstrosity called virDomainChrGetDomainPtrsInternal and friends and mabye produce a saner code that will be easier to introspect both for humans and compilers.
I guess that's matter of taste. I find the current code compact and easy to read. What would be easier is if we dropped def->consoles, def->serials, def->parallels ... in favour of one array, say def->chr where all of those are merged together. But that's a bigger portion of work, isn't it.
Michal Privoznik (2): apibuild: Substitute only pure number tokens virSocketAddrIsPrivate: Work on 32bits platforms
ACK to those two. This time I verified that I've applied the patch prior to trying to break it.
Thanks, pushed. Michal