On Thu, Jan 28, 2021 at 06:28:58PM +0100, Tim Düsterhus wrote: > Willy, > > Am 28.01.21 um 17:27 schrieb Willy Tarreau: > > On Thu, Jan 28, 2021 at 09:19:03PM +0500, ???? ??????? wrote: > >> something went wrong > >> > >> https://github.com/haproxy/haproxy/runs/1781046618?check_suite_focus=true#step:4:40 > > > > Maybe the coverity build scripts use the single quote as an internal > > It appears as if Coverity does not take quoting into account at all, > instead simply splitting the arguments at whitespace. This is dumb of > course.
You're probably right, indeed. > > delimiter. I must say I really hate this development model consisting > > in committing just to see if that fixes, then fix again, not only it's > > ugly, but it wastes everyone's time. > > Sorry about this. I generally attempt to not send any untested patches, > instead testing with my GitHub fork of HAProxy. I think I usually > succeed with this ... I know and am not blaming you for this at all. I'm just saying that it's the limit of this model "commit and pray". It reminds me too much of the old horrible development cycle we used to have 10 years ago for our ALOHA appliance which required to commit into SVN before building, resulting in 95% of SVN commits having "update" for sole message, and I'm really careful about not bringing these horrors back. > Unfortunately in this case I was unable to test the patch due to the > lack of access to Coverity, so I sent a "best effort" patch that I > assumed would work after Ilya reported that apparently the BUG_ON() was > not working. I know, don't worry. The model is absurd, not what you did. Imagine the day we start to use compilers as a service... The real problem is that all these tools are hardly observable and debuggable. Each time we wasted several hours on Travis it was due to strange behaviors that were not locally reproducible and not easily instrumented remotely. And from time to time we're seeing the same on Coverity for the same reasons. > > Stupid question, doesn't coverity pre-set some defines that could be > > checked in the code instead of constantly having to define extra > > "DEBUG_xxx" stuff to shut it up ? Maybe there's even something for > > the unreachable case by the way. > > Ilya will need to answer this. No problem. It was just an idea so that we can hope to more easily shut it up when such issues come up in the future anyway. Cheers, Willy
