(two corrections, I noticed I did not wrote what I meant, previously) On Sat, Feb 25, 2017 at 8:47 AM, Jiri Danek <[email protected]> wrote:
> The mkstemp() thing is supposed to be an issue only with old glibc on > Linux and on AIX and HP-UX. I googled out that there is a good way to set > umask() for the process, because there is JVM running in it too... So > probably ignore. I meant to say "there is _not_ a good way" On Sat, Feb 25, 2017 at 9:11 AM, Jiri Danek <[email protected]> wrote: > Hello, let me sent one more e-mail in this thread ;P Trying to evaluate > what Coverity found, I opened some Jiras where I understood the problem and > it seemed real, and marked the Coverity find as ignored if . I meant to add "marked as ignored if I was reasonably sure that it is intended". You can filter for this on the Coverity web. There were two issues where it was obvious, and one where I think it is not worth fixing: 204 if (futureListener != null) { 205 // TODO BEFORE MERGE: (is null a good option here?) CID 1409858 (#1 of 1): Explicit null dereferenced (FORWARD_NULL) var_deref_model: Passing null to operationComplete, which dereferences it. (The virtual call resolves to io.netty.channel.embedded.EmbeddedChannel.1 .operationComplete.) 206 futureListener.operationComplete(null); 207 } It fails the substitution principle (most of ChannelFutureListener implementations that exist in Netty would fail with NPE there), but what I found by searching in Intellij is that only null or a barebone ChannelFutureListener that does work are ever passed there. To fix it, one would have to implement proper instance of ChannelFuture to pass as argument instead of that null, which would be 100 lines of essentially fluff, and there would still be problem because the future needs to be able to return non-null instance of Channel, of which there isn't any. I simply cannot think of reasonable fix. And it seems to be working all right. Maybe change the comment ;) -- Jiří Daněk
